ai-modules epistemic report

Type: agentic-system-analyses/types/agentic-system-epistemic-report.md

Source-and-claim boundary

Question: what content can become evidence for later behavior, what checks license that reliance, and where do retention or continuation outrun acceptance? Scope and host exclusions are see boundary; evidence identities/layers are see Source register. SRC-1 is the only source. Direct inspection below supplies implementation or doctrine/design, not observed execution. Runtime findings remain provisional; no defective supplied identity or necessary supersession was found. Its combined routes remain valid containers, with material parts declared here.

Public/design claims are see RT-CLM-persistent-context, EPI-CLM-spr-recoverability, EPI-CLM-shaper-safety, and EPI-CLM-editor-fidelity. Scope is not a universal truth engine: source-backed pages, task completion and regression measurements require different evaluators and licenses.

Entry point/operation	Source path	Coverage and records	Remaining limit / conclusion prevented
Plugin deployment, helper dispatch, charter hook	SRC-1 deployment/deployment.sh; plugins/ai_dev/skills/agent_spinner/SKILL.md; plugins/ai_dev/hooks/charter_guardrail.sh	Supplied runtime: RT-RTE-deploy-artifacts, RT-RTE-helper-dispatch, RT-RTE-charter-protect	Not independently reread here; operational permission/installation is not truth approval.
Direct wiki ingest/query/archive	SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md	Direct reads capture/write, procedure, query, archive and lint blocks; EPI-RTE-wiki-compose, EPI-RTE-wiki-retain	No paired claim/source execution instances; preservation/entailment remains indeterminate.
Import/wrapup selection	SRC-1 plugins/knowledge_management/skills/wiki_import/SKILL.md; plugins/knowledge_management/skills/wiki_wrapup/SKILL.md	Supplied RT-RTE-wiki-triage	Human selection is not demonstrated truth acceptance.
Wiki maintenance	SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md	Direct remediation contract, objectives, contradiction/provenance/confidence and fix passages; EPI-RTE-wiki-critic, EPI-RTE-wiki-dispute	Remaining full working-set/cache and other repair internals not comprehensively read; no complete maintenance guarantee.
Wiki lint/hash CLI	SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py; plugins/knowledge_management/skills/wiki/scripts/compute_sha256.py	Direct Lint parsing/matching/partition, quality/main/report dispatch; hash split/update/process/main through write loop; EPI-RTE-lint-check, EPI-RTE-lint-dispose, EPI-RTE-lint-acknowledge	Not every checker body inspected. Hash matches body, not authenticity or endorsement; refresh can replace historical mismatch evidence.
Task implementation/readiness helpers	SRC-1 plugins/ai_dev/skills/task_implement/SKILL.md; plugins/ai_dev/skills/task_auto_check/SKILL.md	Supplied RT-RTE-task-implement, RT-RTE-readiness-repair	No independent run witness; requested proposal veto/rollback does not prove exercise.
Task audit/finish	SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md; plugins/ai_dev/skills/task_finish/SKILL.md	Direct audit full procedure and finish verification consumer; EPI-RTE-task-audit, EPI-RTE-task-stamp	Base archive internals not reread; no completion accuracy or observed close-out.
Wiki layer-2 evaluator	SRC-1 tests/wiki/layer2/normalize.py; tests/wiki/layer2/grade.py; tests/wiki/layer2/aggregate.py; tests/wiki/layer2/run.py	Direct full normalize/grade/aggregate, caller final dispatch; EPI-RTE-report-normalize, EPI-RTE-scenario-check, EPI-RTE-scenario-dispose, EPI-RTE-benchmark-rollup; supplied launch RT-RTE-evaluator-worker	Scenario specifications/fixtures and retained run traces not examined; evaluator capacity, not actual pass rate or causal efficacy.
SPR, summary, humanizer	SRC-1 plugins/knowledge_management/skills/spr/SKILL.md; plugins/knowledge_management/skills/executive_summary/SKILL.md; plugins/ai_editorial/skills/language_humanizer/SKILL.md	Direct complete skill bodies; EPI-RTE-text-reshape	Their grader/judge suites not inspected; no semantic preservation or reconstruction benefit established.
Other shipped model skills/evaluators	SRC-1 plugins/ai_dev/skills/; tests/	Tree/search inventory only: git review/commit, guardrail audit, instruction writing/doctor, portability, task tree repair/select/create, changelog, formatting and non-wiki evaluators remain unassessed	Available but uninspected. They prevent a system-complete negative about additional criticism, acceptance or learning routes.
Retained local task/wiki instances, standing authoring instructions and styles	SRC-1 tasks/; wiki/; AGENTS.md; TESTING.md; styles/	Initial boundary/runtime coverage only	No claim that retained instance presence proves production/check/acceptance provenance; no corpus-wide theory verdict.

No target/test/provider execution occurred. Missing paired semantic outputs prevents transformation classification beyond the stated bounds. Missing invoked helper traces prevents operational veto/rollback claims. Missing provenance-complete worker traces prevents observed confinement or benchmark judgments. Missing controlled interventions prevents component-effect, transfer and improvement attribution. External model internals remain see RT-CMP-host-model: no local evidence of parametric updating or immutable provider version. Uninspected operations are not access gaps or absences.

Epistemic-object inventory

Object ID	Candidate truth-apt content or none	Claimed role	Evidence source ID and local anchor	Gap/limit
EPI-OBJ-raw-sidecars	see EPI-OBJ-raw-sidecars	Raw source sidecars	SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md	see EPI-OBJ-raw-sidecars
EPI-OBJ-wiki-pages	see EPI-OBJ-wiki-pages	Wiki pages	SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md	see EPI-OBJ-wiki-pages
EPI-OBJ-lint-findings	see EPI-OBJ-lint-findings	Linter Issue results	SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py	see EPI-OBJ-lint-findings
EPI-OBJ-lint-settings	see EPI-OBJ-lint-settings	SCHEMA lint decisions	SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py	see EPI-OBJ-lint-settings
EPI-OBJ-audit-verdict	see EPI-OBJ-audit-verdict	Task audit completion verdict	SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md	see EPI-OBJ-audit-verdict
EPI-OBJ-test-report	see EPI-OBJ-test-report	Wiki evaluation evidence	SRC-1 tests/wiki/layer2/grade.py	see EPI-OBJ-test-report
EPI-OBJ-text-output	see EPI-OBJ-text-output	Standalone transformed documents	SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md	see EPI-OBJ-text-output
RT-OBJ-relay-files	see RT-OBJ-relay-files	Task repair relay	SRC-1 plugins/ai_dev/skills/task_auto_check/SKILL.md	see RT-RTE-readiness-repair

Authority-route ledger

Ledger fields owned by records say see; architectural status and content/update relation are local assessments. Doctrine-only means the requested model procedure is inspected, not that no executable script assists it. Checking and disposition have separate rows; retention is not labeled lifecycle integration.

Route ID: EPI-RTE-wiki-compose Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: truth-apt transformation: indeterminate Transition or check target: see EPI-RTE-wiki-compose Evaluator/condition and domain: see EPI-RTE-wiki-compose Activation and timing: see EPI-RTE-wiki-compose Possible or observed result: see EPI-RTE-wiki-compose Implemented force: see EPI-RTE-wiki-compose Epistemic authority and scope: see EPI-RTE-wiki-compose Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-compose Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-compose Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md Claim IDs or none: RT-CLM-persistent-context Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-compose

Route ID: EPI-RTE-wiki-retain Route function: retention Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: no content change Transition or check target: see EPI-RTE-wiki-retain Evaluator/condition and domain: see EPI-RTE-wiki-retain Activation and timing: see EPI-RTE-wiki-retain Possible or observed result: see EPI-RTE-wiki-retain Implemented force: see EPI-RTE-wiki-retain Epistemic authority and scope: see EPI-RTE-wiki-retain Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-retain Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-retain Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md Claim IDs or none: RT-CLM-persistent-context Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-retain

Route ID: EPI-RTE-wiki-critic Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: no content change Transition or check target: see EPI-RTE-wiki-critic Evaluator/condition and domain: see EPI-RTE-wiki-critic Activation and timing: see EPI-RTE-wiki-critic Possible or observed result: see EPI-RTE-wiki-critic Implemented force: see EPI-RTE-wiki-critic Epistemic authority and scope: see EPI-RTE-wiki-critic Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-critic Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-critic Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md Claim IDs or none: EPI-CLM-shaper-safety Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-critic

Route ID: EPI-RTE-wiki-dispute Route function: disposition/acceptance Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: non-truth-apt policy/content update: contested metadata Transition or check target: see EPI-RTE-wiki-dispute Evaluator/condition and domain: see EPI-RTE-wiki-dispute Activation and timing: see EPI-RTE-wiki-dispute Possible or observed result: see EPI-RTE-wiki-dispute Implemented force: see EPI-RTE-wiki-dispute Epistemic authority and scope: see EPI-RTE-wiki-dispute Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-dispute Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-dispute Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md Claim IDs or none: EPI-CLM-shaper-safety Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-dispute

Route ID: EPI-RTE-lint-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-OBJ-lint-findings Content/update relation: no content change Transition or check target: see EPI-RTE-lint-check Evaluator/condition and domain: see EPI-RTE-lint-check Activation and timing: see EPI-RTE-lint-check Possible or observed result: see EPI-RTE-lint-check Implemented force: see EPI-RTE-lint-check Epistemic authority and scope: see EPI-RTE-lint-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-lint-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-lint-check Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-lint-check

Route ID: EPI-RTE-lint-dispose Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: see EPI-OBJ-lint-findings Content/update relation: no content change Transition or check target: see EPI-RTE-lint-dispose Evaluator/condition and domain: see EPI-RTE-lint-dispose Activation and timing: see EPI-RTE-lint-dispose Possible or observed result: see EPI-RTE-lint-dispose Implemented force: see EPI-RTE-lint-dispose Epistemic authority and scope: see EPI-RTE-lint-dispose Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-lint-dispose Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-lint-dispose Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-lint-dispose

Route ID: EPI-RTE-lint-acknowledge Route function: behavior/policy adaptation Architectural status: implemented Object/candidate ID: see EPI-OBJ-lint-settings Content/update relation: non-truth-apt policy/content update: finding suppression Transition or check target: see EPI-RTE-lint-acknowledge Evaluator/condition and domain: see EPI-RTE-lint-acknowledge Activation and timing: see EPI-RTE-lint-acknowledge Possible or observed result: see EPI-RTE-lint-acknowledge Implemented force: see EPI-RTE-lint-acknowledge Epistemic authority and scope: see EPI-RTE-lint-acknowledge Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-lint-acknowledge Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-lint-acknowledge Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-lint-acknowledge

Route ID: EPI-RTE-task-audit Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-audit-verdict Content/update relation: no content change Transition or check target: see EPI-RTE-task-audit Evaluator/condition and domain: see EPI-RTE-task-audit Activation and timing: see EPI-RTE-task-audit Possible or observed result: see EPI-RTE-task-audit Implemented force: see EPI-RTE-task-audit Epistemic authority and scope: see EPI-RTE-task-audit Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-task-audit Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-task-audit Evidence source ID and local anchor: SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-task-audit

Route ID: EPI-RTE-task-stamp Route function: disposition/acceptance Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-audit-verdict Content/update relation: non-truth-apt policy/content update: lifecycle status Transition or check target: see EPI-RTE-task-stamp Evaluator/condition and domain: see EPI-RTE-task-stamp Activation and timing: see EPI-RTE-task-stamp Possible or observed result: see EPI-RTE-task-stamp Implemented force: see EPI-RTE-task-stamp Epistemic authority and scope: see EPI-RTE-task-stamp Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-task-stamp Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-task-stamp Evidence source ID and local anchor: SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-task-stamp

Route ID: EPI-RTE-report-normalize Route function: content transformation Architectural status: implemented Object/candidate ID: see EPI-OBJ-test-report Content/update relation: truth-apt transformation: non-ampliative reshaping Transition or check target: see EPI-RTE-report-normalize Evaluator/condition and domain: see EPI-RTE-report-normalize Activation and timing: see EPI-RTE-report-normalize Possible or observed result: see EPI-RTE-report-normalize Implemented force: see EPI-RTE-report-normalize Epistemic authority and scope: see EPI-RTE-report-normalize Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-report-normalize Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-report-normalize Evidence source ID and local anchor: SRC-1 tests/wiki/layer2/normalize.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-report-normalize

Route ID: EPI-RTE-scenario-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-OBJ-test-report Content/update relation: truth-apt transformation: entailed derivation Transition or check target: see EPI-RTE-scenario-check Evaluator/condition and domain: see EPI-RTE-scenario-check Activation and timing: see EPI-RTE-scenario-check Possible or observed result: see EPI-RTE-scenario-check Implemented force: see EPI-RTE-scenario-check Epistemic authority and scope: see EPI-RTE-scenario-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-scenario-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-scenario-check Evidence source ID and local anchor: SRC-1 tests/wiki/layer2/grade.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-scenario-check

Route ID: EPI-RTE-scenario-dispose Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: see EPI-OBJ-test-report Content/update relation: no content change Transition or check target: see EPI-RTE-scenario-dispose Evaluator/condition and domain: see EPI-RTE-scenario-dispose Activation and timing: see EPI-RTE-scenario-dispose Possible or observed result: see EPI-RTE-scenario-dispose Implemented force: see EPI-RTE-scenario-dispose Epistemic authority and scope: see EPI-RTE-scenario-dispose Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-scenario-dispose Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-scenario-dispose Evidence source ID and local anchor: SRC-1 tests/wiki/layer2/grade.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-scenario-dispose

Route ID: EPI-RTE-benchmark-rollup Route function: operational admission/selection/consumption Architectural status: implemented Object/candidate ID: see EPI-OBJ-test-report Content/update relation: truth-apt transformation: entailed derivation Transition or check target: see EPI-RTE-benchmark-rollup Evaluator/condition and domain: see EPI-RTE-benchmark-rollup Activation and timing: see EPI-RTE-benchmark-rollup Possible or observed result: see EPI-RTE-benchmark-rollup Implemented force: see EPI-RTE-benchmark-rollup Epistemic authority and scope: see EPI-RTE-benchmark-rollup Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-benchmark-rollup Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-benchmark-rollup Evidence source ID and local anchor: SRC-1 tests/wiki/layer2/aggregate.py Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-benchmark-rollup

Route ID: EPI-RTE-text-reshape Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-text-output Content/update relation: truth-apt transformation: indeterminate Transition or check target: see EPI-RTE-text-reshape Evaluator/condition and domain: see EPI-RTE-text-reshape Activation and timing: see EPI-RTE-text-reshape Possible or observed result: see EPI-RTE-text-reshape Implemented force: see EPI-RTE-text-reshape Epistemic authority and scope: see EPI-RTE-text-reshape Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-text-reshape Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-text-reshape Evidence source ID and local anchor: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md Claim IDs or none: EPI-CLM-editor-fidelity Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-text-reshape

Route ID: EPI-RTE-raw-capture Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-raw-sidecars Content/update relation: truth-apt transformation: acquisition/import Transition or check target: see EPI-RTE-raw-capture Evaluator/condition and domain: see EPI-RTE-raw-capture Activation and timing: see EPI-RTE-raw-capture Possible or observed result: see EPI-RTE-raw-capture Implemented force: see EPI-RTE-raw-capture Epistemic authority and scope: see EPI-RTE-raw-capture Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-raw-capture Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-raw-capture Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-raw-capture

Route ID: EPI-RTE-wiki-remediate Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: truth-apt transformation: indeterminate Transition or check target: see EPI-RTE-wiki-remediate Evaluator/condition and domain: see EPI-RTE-wiki-remediate Activation and timing: see EPI-RTE-wiki-remediate Possible or observed result: see EPI-RTE-wiki-remediate Implemented force: see EPI-RTE-wiki-remediate Epistemic authority and scope: see EPI-RTE-wiki-remediate Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-remediate Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-remediate Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-remediate

Route ID: EPI-RTE-wiki-archive Route function: operational admission/selection/consumption Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-wiki-pages Content/update relation: no content change Transition or check target: see EPI-RTE-wiki-archive Evaluator/condition and domain: see EPI-RTE-wiki-archive Activation and timing: see EPI-RTE-wiki-archive Possible or observed result: see EPI-RTE-wiki-archive Implemented force: see EPI-RTE-wiki-archive Epistemic authority and scope: see EPI-RTE-wiki-archive Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-wiki-archive Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-wiki-archive Evidence source ID and local anchor: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-wiki-archive

Route ID: EPI-RTE-text-check Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-OBJ-text-output Content/update relation: no content change Transition or check target: see EPI-RTE-text-check Evaluator/condition and domain: see EPI-RTE-text-check Activation and timing: see EPI-RTE-text-check Possible or observed result: see EPI-RTE-text-check Implemented force: see EPI-RTE-text-check Epistemic authority and scope: see EPI-RTE-text-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-text-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-text-check Evidence source ID and local anchor: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md Claim IDs or none: none Mismatch marker or none: none; limits are record-local Gap/limit: see EPI-RTE-text-check

Route ID: RT-RTE-wiki-triage Route function: operational admission/selection/consumption Architectural status: doctrine only Content/update relation: no content change Object/candidate ID: see RT-RTE-wiki-triage Transition or check target: see RT-RTE-wiki-triage Evaluator/condition and domain: see RT-RTE-wiki-triage Activation and timing: see RT-RTE-wiki-triage Possible or observed result: see RT-RTE-wiki-triage Implemented force: see RT-RTE-wiki-triage Epistemic authority and scope: see RT-RTE-wiki-triage Operational authority: behavior permitted, blocked, or changed: see RT-RTE-wiki-triage Behavioral-authority path: consumer, channel, force, horizon: see RT-RTE-wiki-triage Evidence source ID and local anchor: see RT-RTE-wiki-triage Claim IDs or none: see RT-RTE-wiki-triage Mismatch marker or none: see RT-RTE-wiki-triage Gap/limit: see RT-RTE-wiki-triage

Route ID: RT-RTE-readiness-repair Route function: disposition/acceptance Architectural status: doctrine only Content/update relation: non-truth-apt policy/content update: permitted task repair Object/candidate ID: see RT-RTE-readiness-repair Transition or check target: see RT-RTE-readiness-repair Evaluator/condition and domain: see RT-RTE-readiness-repair Activation and timing: see RT-RTE-readiness-repair Possible or observed result: see RT-RTE-readiness-repair Implemented force: see RT-RTE-readiness-repair Epistemic authority and scope: see RT-RTE-readiness-repair Operational authority: behavior permitted, blocked, or changed: see RT-RTE-readiness-repair Behavioral-authority path: consumer, channel, force, horizon: see RT-RTE-readiness-repair Evidence source ID and local anchor: see RT-RTE-readiness-repair Claim IDs or none: see RT-RTE-readiness-repair Mismatch marker or none: see RT-RTE-readiness-repair Gap/limit: see RT-RTE-readiness-repair

Route ID: RT-RTE-wiki-consume Route function: operational admission/selection/consumption Architectural status: doctrine only Content/update relation: no content change Object/candidate ID: see RT-RTE-wiki-consume Transition or check target: see RT-RTE-wiki-consume Evaluator/condition and domain: see RT-RTE-wiki-consume Activation and timing: see RT-RTE-wiki-consume Possible or observed result: see RT-RTE-wiki-consume Implemented force: see RT-RTE-wiki-consume Epistemic authority and scope: see RT-RTE-wiki-consume Operational authority: behavior permitted, blocked, or changed: see RT-RTE-wiki-consume Behavioral-authority path: consumer, channel, force, horizon: see RT-RTE-wiki-consume Evidence source ID and local anchor: see RT-RTE-wiki-consume Claim IDs or none: see RT-RTE-wiki-consume Mismatch marker or none: see RT-RTE-wiki-consume Gap/limit: see RT-RTE-wiki-consume

System-claim versus route comparison

Claim ID	Claimed operation or warrant	Claim source ID/anchor and layer	Doctrine/design support	Implemented route IDs	Observed-run support	Causal support and design limits	Supported conclusion	Mismatch/unknown
RT-CLM-persistent-context	see record	SRC-1 README.md, doctrine/design	see record	EPI-RTE-lint-check; model retention/consumption is doctrine	No trace assessed	No controlled contrast; retention does not show improved accuracy	Persistent context is afforded; exercised benefit remains unknown	Availability is not acceptance
EPI-CLM-spr-recoverability	see record	SRC-1 plugins/knowledge_management/skills/spr/SKILL.md, doctrine/design	see record	None for reconstruction; EPI-RTE-text-reshape is doctrine	None assessed	Self-rating/coverage lacks independent receiver reconstruction	Explicit source-bound compression contract	Recoverability and parametric activation mechanism unverified
EPI-CLM-shaper-safety	see record	SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md, doctrine/design	see record	EPI-RTE-lint-check, EPI-RTE-lint-dispose cover narrower mechanical checks	None assessed	Safe/lossless is model judgment, not executable invariant	Ambiguities are requested to be surfaced and conflicts not auto-resolved	Universal unattended safety unsupported
EPI-CLM-editor-fidelity	see record	SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md, doctrine/design	see record	EPI-RTE-text-reshape is doctrine	None assessed	Same model builds ledger and verifies text; no independent truth check	Explicit content-preservation procedure	Fidelity/comprehension benefit unknown

Bounded conclusion

The inspected wiki routes acquire external/session content and deliver retained claims to later answer/write decisions; see RT-RTE-wiki-triage, EPI-RTE-wiki-compose, EPI-RTE-wiki-retain, RT-BAP-wiki-guidance. Source attribution and body hashes preserve pointers/self-consistency, not source truth; EPI-RTE-raw-capture. Requested deterministic repair and active archive withdrawal differ from accepted-claim integration; EPI-RTE-wiki-remediate, EPI-RTE-wiki-archive. Archive changes active navigation but does not establish complete exclusion from the query's recursive Markdown search. Semantic warrant remains unknown. First synthesis is explicitly retained while user validation is pending; confidence is not a recorded evidence-consuming truth acceptance. Cross-page reasoning supplies candidate production, but no inspected input/output establishes entailment or non-entailment. Thus neither accepted ampliative knowledge nor its lifecycle integration is established by this witness.

Checks have different domains and forces. Wiki lint establishes only mechanical predicates and its exit permits warnings; EPI-RTE-lint-check, EPI-RTE-lint-dispose. Semantic conflict checking requests that both positions survive with human-facing dispute signals, not that retrieval is blocked; EPI-RTE-wiki-critic, EPI-RTE-wiki-dispute. Intent-preserving task repair has proposed computational veto and rollback see RT-RTE-readiness-repair; it is not a universal pre-write gate.

A stronger named acceptance criterion exists for task completion: code/tests plus every body/Acceptance item and design signal precede audited stamping. Later close-out consumes that status, rechecking on concrete drift evidence; EPI-RTE-task-audit, EPI-RTE-task-stamp. This licenses task-specific completion reliance under the protocol, not general factual truth, test completeness or observed success. Initiating the audit is human control of the request; the model determines the verdict/stamp.

The wiki evaluator implements scoped artifact checking, final run-completion gates and regression reporting, not product-change acceptance or causal explanation. Normalization omits some response context; report predicates are self-report checks. Assertion averages and final passed are distinct and remain distinct in aggregation; EPI-RTE-report-normalize, EPI-RTE-scenario-check, EPI-RTE-scenario-dispose, EPI-RTE-benchmark-rollup. These deterministic deductions are warranted conditional on input artifacts and inspected fixed interpretation, not on claims that a worker actually complied.

SCHEMA lint acknowledgements directly adapt future report behavior without a truth-apt content route: info matches disappear from live counts, with normal-mode acknowledgement and quiet-mode omission; EPI-RTE-lint-acknowledge. This is policy-input consumption; the inspected shaper can author size exceptions, so a mechanically accepted finding need not evidence human review. It neither proves factual endorsement nor suppresses blocking issues.

Standalone compression/editing provides fidelity contracts and requested self-checks, not demonstrated semantic preservation or independent acceptance; EPI-RTE-text-reshape, EPI-RTE-text-check. Parameters remain opaque see RT-CMP-host-model. These positives leave learning benefit, exercised reflection, theory-specific content criticism/iteration and self-improvement attribution independently unestablished: no selected retained theory instance or operation-responsive method revision and later exercised consumer trace was assessed. Computational ownership is supported for bookkeeping and requested task/wiki substeps; it does not imply autonomous reliable knowledge production. Remaining skill/evaluator families prevent a whole-system negative or a single epistemic verdict.

Shared records

Operative objects

EPI-OBJ-raw-sidecars

Label: Raw source sidecars

Part of: RT-RTE-wiki-consume

This object is material state/result carried by the parent route (cross-kind containment), not a replacement for its invocation/consumer progression.

Captured external/session content and origin/body-hash frontmatter. Body content is natural-language source material; origin pointers and hashes are symbolic lineage fields. Markdown filesystem. Truth-apt source assertions remain imported and their truth/fidelity is unknown. Hash checks compare current body bytes with recorded digest, not source endorsement. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md; supporting passages in the route records below.

EPI-OBJ-wiki-pages

Label: Wiki pages

Part of: RT-RTE-wiki-consume

This object is material state/result carried by the parent route (cross-kind containment), not a replacement for its invocation/consumer progression.

Typed page bodies and confidence/contested/checked frontmatter; raw inputs are separately inventoried from synthesized claims. Natural-language assertions and annotations interpreted by model/human; Markdown filesystem. Candidate truth-apt content includes source-attributed facts, comparisons and recommendations with named domain conditions. Metadata records asserted confidence/freshness, not independently verified warrant. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md; supporting passages in the route records below.

EPI-OBJ-lint-findings

Label: Linter Issue results

Part of: RT-RTE-wiki-consume

This object is material state/result carried by the parent route (cross-kind containment), not a replacement for its invocation/consumer progression.

Issue severity/category/path/message and rendered report/exit. Symbolic predicate results in Python memory/stdout; truth-apt over inspected filesystem/schema conformance, not proposition truth. Distinct material result of linter subfunction within the supplied wiki route. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py; supporting passages in the route records below.

EPI-OBJ-lint-settings

Label: SCHEMA lint decisions

Distinct identity: no counterpart in supplied runtime objects after comparison; related task/skill/config containers do not individuate this material content/result.

Unfenced Lint section Accepted finding, Page-check exclusions and Declared boilerplate bullets. Symbolic configuration in Markdown filesystem; no truth-apt content required for their policy effect. They control inspected coverage/reporting rather than licensed factual beliefs. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py; supporting passages in the route records below.

EPI-OBJ-audit-verdict

Label: Task audit completion verdict

Distinct identity: no counterpart in supplied runtime objects after comparison; related task/skill/config containers do not individuate this material content/result.

Task body/Acceptance references, success/gap report and durable audited/implemented status. Natural-language completion claim interpreted by model, lifecycle strings in task Markdown. Truth-apt over one task and current code/tests. Separate downstream audit from runtime implementation/readiness, with different target and consumer. Evidence: SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md; supporting passages in the route records below.

EPI-OBJ-test-report

Label: Wiki evaluation evidence

Part of: RT-RTE-evaluator-worker

This object is material state/result carried by the parent route (cross-kind containment), not a replacement for its invocation/consumer progression.

Delimited TEST REPORT, response, snapshot, timing/provenance, grading and benchmark artifacts. Report assertions are natural language; parsed fields, assertion booleans and statistics are symbolic. Filesystem JSON/Markdown plus Python memory. Split consumption paths in routes: normalization changes delivered evidence, grading checks bounded predicates, final gates change pass classification, aggregation consumes both gated and ungated measures. Evidence: SRC-1 tests/wiki/layer2/grade.py; supporting passages in the route records below.

EPI-OBJ-text-output

Label: Standalone transformed documents

Distinct identity: no counterpart in supplied runtime objects after comparison; related task/skill/config containers do not individuate this material content/result.

Source facts/drafts transformed by spr, executive_summary or language_humanizer. Natural-language content in host conversation or caller-chosen document; no specified dedicated store. Candidate truth-apt assertions mixed with stylistic policy; preservation/reconstruction remains untested. Evidence: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md; supporting passages in the route records below.

Routes

EPI-RTE-wiki-compose

Label: Wiki claim composition

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Foreground model reads captured sources or selected pages, authors sourced claims or cross-page reasoning, and annotates confidence. Target is the relationship between source content and page claims; semantic preservation, entailment and non-entailment are not individually inspectable without paired inputs/outputs. Source warrant is unknown: capture and attribution do not authenticate source truth. Page thresholds and confidence are selection heuristics, not proof. Open ingest/query; user directs the request, model proposes and on the direct hub/query path decides admission. Import/wrapup selections are different, see RT-RTE-wiki-triage.

Endpoints/progression: selected inputs → wiki claim composition → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md.

• Immediate return: Cited answer or proposed/updated page text.
• Later read-back: Wiki pages can reach later query/ingest/maintenance invocations.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Domain relevance; central single-source entity or two-source presence; query filing on three pages or novel cross-page reasoning.
• Invalidation or expiry: Ordinary correction, contested flag or full supersession/archive; no automatic pending-validation expiry.
• Activation or effect: Requested new truth-apt candidates; no exercise or acceptance witnessed.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Write medium for a first synthesis however many pages it draws on: --- plugins/knowledge_management/skills/wiki/SKILL.md

EPI-RTE-wiki-retain

Label: Wiki candidate retention

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Foreground model files first synthesis before user validation and updates index/log. This grants availability to later readers, not an evidence-consuming acceptance decision against a truth criterion. Channel is page files plus navigation; force is later evidence supply across sessions. Human can correct next turn; ordinary updates are recovery, not atomic rollback. No accepted-claim lifecycle integration is established by this retention witness.

Endpoints/progression: selected inputs → wiki candidate retention → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md.

• Immediate return: Page path and filing decision.
• Later read-back: Retained candidate pages selected by subsequent wiki requests.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Three-page-or-novel-reasoning trigger; a skipped eligible answer must state a reason.
• Invalidation or expiry: User correction or archive; pending synthesis has no specified expiry.
• Activation or effect: Requested persistence and future availability before validation.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

File the first synthesis while validation is pending. --- plugins/knowledge_management/skills/wiki/SKILL.md

EPI-RTE-wiki-critic

Label: Wiki semantic criticism

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Maintenance model checks same-subject page pairs for factual, definitional, scope, recency or recommendation disagreements; it also checks provenance, confidence and anatomy. Target is cross-page consistency and source/page agreement, not independent truth of either source. Open user-triggered corpus maintenance; model identifies issues, human decides genuinely ambiguous conflicts. Reference inputs are pages, sources and schema, not a known-correct semantic oracle.

Endpoints/progression: selected inputs → wiki semantic criticism → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md.

• Immediate return: Issue inventory with both excerpts and disagreement dimension.
• Later read-back: uninspected — no actual retained criticism/consumer trace inspected.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Audit working-set pages and same-subject peers found by inventory search.
• Invalidation or expiry: Reassessment on next audit; dispute remains until human resolution.
• Activation or effect: Requested criticism input to remediation; judgment accuracy unobserved.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Flag the contradiction on both pages so the remediate phase can mark both via the contested-page protocol. Do not pick a winner. --- plugins/knowledge_management/agents/auto_shaper_wiki.md

EPI-RTE-wiki-dispute

Label: Wiki dispute disposition

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Maintenance model consumes contradiction findings, marks both pages contested, lists peers and bumps updated without choosing a winner. Consumer is subsequent audit/query model through page frontmatter plus user through final report. Force is advisory warning, not enforced withdrawal from query. Human owns resolution; body replacement or winner-selection is forbidden to the helper. Safe deterministic fixes are separately authorized only when recoverable without lost meaning or invented values. No per-page human approval invariant follows.

Endpoints/progression: selected inputs → wiki dispute disposition → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md.

• Immediate return: Contested metadata and excerpts for human review.
• Later read-back: Flags and pages can reach later readers; no filter denying their retrieval was established in inspected query instructions.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Detected disagreement between same-subject peers; preexisting contested pages left for human decision.
• Invalidation or expiry: Human ordinary update/supersession; no timer or automatic resolution.
• Activation or effect: Requested changed reliance signal, not epistemic rejection of both claims.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Do not edit either page's body to pick a winner, merge the claims, or hedge the wording. --- plugins/knowledge_management/agents/auto_shaper_wiki.md

EPI-RTE-lint-check

Label: Wiki structural checking

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Python checks frontmatter, enum/date syntax, path existence, links/index, hash self-consistency and quality flags. Target is filesystem/text conformance under schema and configured exclusions. Evaluator uses fixed predicates; reference is inspected text and schema, not a semantic truth oracle. These findings license only those predicates. Contested status is detected, not resolved. Page-walk exclusions and boilerplate declarations condition coverage. Blocking issues are not suppressed by Accepted finding bullets.

Endpoints/progression: selected inputs → wiki structural checking → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-lint-findings and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py.

• Immediate return: Severity-tagged Issue list rendered for operator/model.
• Later read-back: inapplicable — immediate finding computation does not itself retain experience.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: CLI wiki discovery/explicit path; page_dirs gate page checks; raw and scaffold checks have separate walks.
• Invalidation or expiry: Recompute against current files/schema; declared exclusions can omit page subtrees.
• Activation or effect: Executable reports feed linter exit and requested repair. No run witnessed.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

issues.extend(check_quality_signals(wiki)) --- plugins/knowledge_management/skills/wiki/scripts/lint.py

EPI-RTE-lint-dispose

Label: Wiki completion disposition

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Python exit returns 1 exactly when a remaining issue is blocking, otherwise 0. Consumer is caller/process and interpreting wiki workflow. Executable force is exit status; prose requests fixes before declaring complete, not rollback or write prevention. Epistemic license is no reported blocking findings within configured checks, not no warnings or true claims. Broad shaper adds a stricter protocol bar: no warns except contested dispute signals. A clean exit cannot substitute for that bar.

Endpoints/progression: selected inputs → wiki completion disposition → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-lint-findings and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py.

• Immediate return: Exit 0 or 1 plus report.
• Later read-back: inapplicable — status is a current-call result, not retained memory by this code.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: Any issue of SEV_BLOCKING after acknowledgement partition.
• Invalidation or expiry: Fresh invocation recalculates; writes already happened before ordinary lint.
• Activation or effect: Programmatic failure signal and requested completion hold; host obedience remains unobserved.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

return 1 if any(i.severity == SEV_BLOCKING for i in issues) else 0 --- plugins/knowledge_management/skills/wiki/scripts/lint.py

EPI-RTE-lint-acknowledge

Label: Lint finding acknowledgement

Distinct identity: no counterpart after comparison with supplied runtime route declarations. This downstream/function-specific operation was not separately declared there.

SCHEMA.md Lint bullets change future reporting: Accepted finding matches only info severity, category/path plus required discriminators. partition_accepted removes matched findings from live counts; normal report retains an acknowledged channel, quiet hides it. Consumer is linter and maintenance model; force is enforcing report classification, then advisory no-action guidance across later audits. The shaper may author size acceptance for a deliberately large synthesis: computational admission, not guaranteed human review despite code comments describing reviewed decisions. Fixed matching does not verify who reviewed a bullet or why. Criterion is intentional soft-cap/style exception for report use, not truth acceptance. Page-check exclusions skip entire trees; Declared boilerplate changes scaffold checking, separately from Accepted finding.

Endpoints/progression: selected inputs → lint finding acknowledgement → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-lint-settings and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/scripts/lint.py.

• Immediate return: Live/acknowledged lists and changed counts.
• Later read-back: Durable SCHEMA bullets are read by later linter invocations.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: Unfenced Lint section, exact category/path and line/message where required; info severity only.
• Invalidation or expiry: Remove/update bullet or change matching discriminator; path-only size acceptance survives moving size counter.
• Activation or effect: Implemented trace of policy input to repeated reporting; no observed adaptation benefit.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

if issue.severity != SEV_INFO: return False --- plugins/knowledge_management/skills/wiki/scripts/lint.py

EPI-RTE-task-audit

Label: Task completion evidence

Distinct identity: no counterpart after comparison with supplied runtime route declarations. This downstream/function-specific operation was not separately declared there.

Model checks each task body/Acceptance item against actual code, required passing tests and design-extended signal, running required suites or consuming qualifying recent expensive results. Target is completion of a named task in its current codebase, not universal program correctness or transfer. Open user-triggered audit; task/tests supply criterion/reference outcomes, not model judgment as oracle. The evaluator attributes failures to audited work versus unrelated preexisting failures. Model reports gaps or full compliance; user owns later close-out and can reject its judgment. Distinct downstream check omitted from runtime detailed account, not a contradiction of its partial coverage.

Endpoints/progression: selected inputs → task completion evidence → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-audit-verdict and the described input/result/consumer chain. Evidence: SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md.

• Immediate return: Requirement/expected/actual/minimum-fix gaps or exact success signal.
• Later read-back: Verdict/status reaches later task_finish; actual exercise uninspected.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: User-named implemented/audited task or archived finished drift recheck; decline still-planned task.
• Invalidation or expiry: Fresh code changes invalidate prior audit reliance; cross-session audit inherits no verification run.
• Activation or effect: Requested evidence production for completion; no suite ran here.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Trust the code, not the prose. --- plugins/ai_dev/skills/task_audit/SKILL.md

EPI-RTE-task-stamp

Label: Task audit disposition

Distinct identity: no counterpart after comparison with supplied runtime route declarations. This downstream/function-specific operation was not separately declared there.

Model consumes full-compliance verdict: stamp implemented task audited only when every body item, Acceptance check, required test and design signal is confirmed. Gaps on audited task reopen to implemented; archived finished recheck does not erase finished. Consumer task_finish trusts audited by default unless concrete code-change evidence demands re-verification; channel durable task frontmatter, force protocol permission to close, horizon task lifecycle across sessions. This is a named evidence-consuming acceptance criterion for task completion, not acceptance of general wiki conjectures. Task content and status persist; repairs go to task_implement/new task, not automatic rollback of code. Computational audit admits the stamp; initiating user request is not per-verdict human approval.

Endpoints/progression: selected inputs → task audit disposition → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-audit-verdict and the described input/result/consumer chain. Evidence: SRC-1 plugins/ai_dev/skills/task_audit/SKILL.md.

• Immediate return: Audited or reopened status and handoff.
• Later read-back: Later close-out reads audited stamp and design-extended; source plugins/ai_dev/skills/task_finish/SKILL.md.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Clean current implemented task admits audited; fixable gaps on audited reopen.
• Invalidation or expiry: Concrete evidence code changed triggers re-verification; historical archive status preserved.
• Activation or effect: Requested permission to close task; no observed closure or correctness.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

on a gap verdict over an audited task, it re-opens the task to status: implemented and bumps updated --- plugins/ai_dev/skills/task_audit/SKILL.md

EPI-RTE-report-normalize

Label: Evaluation report normalization

Part of: RT-RTE-evaluator-worker

Separate material function within that container; no supersession needed.

Python extracts first delimited TEST REPORT into missing/empty report.md and removes configured response preambles. Target content is worker claims about its actions, not verified actions. Concrete deduction: input response "The harness blocked the file write.\n\n==== TEST REPORT ====\nlinted: yes\n==== END REPORT ====\n" with absent report yields extracted report carrying linted: yes, while first NOISE_PATTERNS regex removes the opening line from cleaned response. The claim survives; the context that writing was blocked does not. If report.md already nonempty it remains preferred by grader even if response differs. This is deterministic selection/reshaping with degraded context lineage, not an external-source truth certificate. Normalization failure is best-effort in caller (check=False).

Endpoints/progression: selected inputs → evaluation report normalization → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-test-report and the described input/result/consumer chain. Evidence: SRC-1 tests/wiki/layer2/normalize.py.

• Immediate return: Recovered report.md, edited response.txt and action messages.
• Later read-back: inapplicable — these are current evaluation artifacts; later grading within same pipeline is not accumulated-experience read-back.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: Known scenario pass directories; response exists; first matched report block; missing/empty report only.
• Invalidation or expiry: Existing report is left untouched; no original-response backup in inspected normalizer.
• Activation or effect: Implemented transformed evidence supply to grader/renderer, not semantic validation.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

report.write_text(block + "\n") --- tests/wiki/layer2/normalize.py

EPI-RTE-scenario-check

Label: Scenario assertion evaluation

Part of: RT-RTE-evaluator-worker

Separate material function within that container; no supersession needed.

Python evaluates scenario assertions against parsed report, cleaned response, snapshots and timing provenance. Target domain is named assertions: report equality proves what was reported; file/hash predicates prove scoped artifact properties. file_exists_and_changed only checks existence and nonzero size, not baseline change. no_files_outside_sandbox approximates using listed paths, not complete observed I/O. A concrete bounded deduction: report field linted: yes with expected yes yields true via lowercased equality, whether or not lint was actually called. Premise is report text, not warranted execution. Oracle is author-supplied assertion expectations; bounded regression evaluation. No claim-truth, mechanism correctness, causal benefit or host confinement follows.

Endpoints/progression: selected inputs → scenario assertion evaluation → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-test-report and the described input/result/consumer chain. Evidence: SRC-1 tests/wiki/layer2/grade.py.

• Immediate return: Per-assertion booleans/evidence and pass_rate.
• Later read-back: inapplicable — current pipeline findings do not automatically train or inform later workers.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: Scenario assertions and allowed/extra report fields; snapshot preferred over removed root.
• Invalidation or expiry: New assertion/fixture changes alter domain; missing fields default empty.
• Activation or effect: Implemented scoped evidence for final grading and benchmarks.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

ok = actual.strip().lower() == assertion["expected"].strip().lower() --- tests/wiki/layer2/grade.py

EPI-RTE-scenario-dispose

Label: Scenario pass disposition

Part of: RT-RTE-evaluator-worker

Separate material function within that container; no supersession needed.

Final passed requires every assertion plus worker_rc==0, nonempty response and integrity/checkout checks when present. Missing integrity dicts default permissive, so absent provenance is not affirmative evidence of clean access. Check result licenses acceptance of this scenario pass under these gates only. Per-assertion pass_rate deliberately remains ungated; grade CLI returns 0 even when a scenario fails, so no process-wide failure follows directly. Aggregator consumes both passed and pass_rate; no automatic instruction/product update is implemented by this pipeline.

Endpoints/progression: selected inputs → scenario pass disposition → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-test-report and the described input/result/consumer chain. Evidence: SRC-1 tests/wiki/layer2/grade.py.

• Immediate return: grading.json/summary passed flag and pass_rate.
• Later read-back: inapplicable — stored evaluation results have no automatic later-worker feedback in inspected path.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: All assertion results and worker/timing conditions.
• Invalidation or expiry: New pass regrades current artifacts; missing provenance does not force rejection.
• Activation or effect: Enforcing boolean classification for benchmark consumer, not a product-admission gate.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

"passed": all_assertions_pass and worker_ok and integrity_ok and checkout_ok, --- tests/wiki/layer2/grade.py

EPI-RTE-benchmark-rollup

Label: Regression benchmark consumption

Part of: RT-RTE-evaluator-worker

Separate material function within that container; no supersession needed.

Aggregator computes statistics over assertion pass_rate, retains per-pass passed and all_passes_clean, and compares prior benchmark for perfect-to-imperfect assertion or clean-to-unclean scenario regressions. Target is those artifact summaries/comparisons; warranted only conditional on grading/timing inputs, not skill improvement. Caller returns aggregate rc: a regression makes exit 1; an already-failing scenario without regression can leave exit 0. Separate clean summary prevents equating high assertion average with all gates passed, but table prominently uses ungated means. Force is CLI signal/report for human maintainer; no automatic production-policy adaptation. Comparison is nonrandomized prior-run regression, not causal component attribution.

Endpoints/progression: selected inputs → regression benchmark consumption → named result/consumer described above. Owner: fixed Python rules; caller/model supplies inputs. Context/state/action effects: see EPI-OBJ-test-report and the described input/result/consumer chain. Evidence: SRC-1 tests/wiki/layer2/aggregate.py.

• Immediate return: benchmark.json/Markdown; regression exit 1 or otherwise 0.
• Later read-back: Previous benchmark can be consumed in a later evaluation run; no worker learning feedback established.
• Delegated visibility: inapplicable — this Python suboperation does not dispatch a helper; enclosing worker route is separate.
• Selection predicate: Known scenario directories and supplied/latest previous benchmark.
• Invalidation or expiry: New artifacts/specs or missing previous data bound comparisons; cross-run conditions not controlled here.
• Activation or effect: Implemented report and caller exit consequence; no benefit claim or execution observation.
• Evidence limits: Inspected implementation and conditional deductions, not target execution or observed benefit.
• implementation conclusion status: wired
• operation conclusion status: uninspected

if ps.get("all_passes_clean") and not s["all_passes_clean"]: --- tests/wiki/layer2/aggregate.py

EPI-RTE-text-reshape

Label: Source-bounded text transformation

Distinct identity: no counterpart after comparison with supplied runtime route declarations. This downstream/function-specific operation was not separately declared there.

SPR, executive_summary and humanizer request source-bounded restructuring without added facts; preservation is a contract, not demonstrated semantic equivalence. Their content-bearing output can contain truth-apt assertions but no paired outputs were assessed. Natural-language self-check differs by target: SPR paragraph coverage/recoverability, summary structure/length and humanizer ledger fidelity. Humanizer explicitly forbids even unstated arithmetic additions; novel fluent phrasing does not establish ampliative conjecture. Open document requests: user supplies material and mode/reader, model authors; ambiguity routes to author. Consumers are user or proposed receiving LLM via delivered text; force advisory/priming, retention unspecified. Distinct from wiki-consume: this referent spans three sibling standalone skills and is not its part.

Endpoints/progression: selected inputs → source-bounded text transformation → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-text-output and the described input/result/consumer chain. Evidence: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md.

• Immediate return: Bullet SPR plus reconstructability integer; prose summary/self-rating; humanizer findings or full text with preservation/open-items notes.
• Later read-back: inapplicable — standalone skills specify delivered text, not a retained cross-session store.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Requested text/mode; SPR matches input language, humanizer names intended reader.
• Invalidation or expiry: Author can correct mode/ambiguities or request revision; no rollback/storage policy specified.
• Activation or effect: Requested altered text/context supply; reconstruction and comprehension effects unobserved.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Take every value from the input. --- plugins/ai_editorial/skills/language_humanizer/SKILL.md

EPI-RTE-raw-capture

Label: Raw source acquisition

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Foreground model fetches/converts/extracts or saves supplied material into raw sidecars before constructing page claims. The raw sidecar is input state carried by the parent wiki route, not itself a synthesized page. Content is acquired, not produced; warrant is unknown because origin fields and conversion do not authenticate source truth or establish extraction fidelity. Human names resource in import, while direct hub may operate in automated contexts. Import can retain raw before page approval; wrapup forbids writes before selection. Hash script only certifies current body self-consistency. Re-ingest check compares candidate against recorded hash before replacement; refresh after mismatch retains the new body/hash, not the old historical mismatch. Guidance is taxonomy/raw-frontmatter contract; no complete transactional recovery.

Endpoints/progression: selected inputs → raw source acquisition → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-raw-sidecars and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md.

• Immediate return: Raw body and origin/hash metadata, then page-writing input.
• Later read-back: Captured material can reach later ingest, query attribution and maintenance.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Named source or supplied paste; resource kind controls raw bucket.
• Invalidation or expiry: Unchanged re-ingest leaves old sidecar; drift replaces body and hash after comparison.
• Activation or effect: Requested import and lineage bookkeeping; actual capture fidelity unobserved.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

URL of an externally-published article → fetch, convert to markdown, save to raw/articles/. --- plugins/knowledge_management/skills/wiki/SKILL.md

EPI-RTE-wiki-remediate

Label: Wiki deterministic repair

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Maintenance model consumes findings and rewrites/splits/relocates content only when safe and deterministic: no lost meaning, invented value or genuine ambiguity resolved. Target check is conformance to schema/anatomy and fidelity to existing meaning; model judgment supplies it, not fixed semantic code. Human receives irreducible cases and can veto resolution; deterministic fixes are computationally admitted without per-fix approval. Source-drift repair re-reads raw against claims; genuine conflicts remain for human. Writes plus per-file report persist; no all-file rollback is established. Claimed non-ampliative reshaping is not witnessed with paired outputs, so relation remains indeterminate. This is not accepted-claim lifecycle integration.

Endpoints/progression: selected inputs → wiki deterministic repair → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md.

• Immediate return: Repaired files and per-file change report, followed by re-lint.
• Later read-back: Maintained pages/scaffold reach later wiki consumers.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Audit findings judged recoverable from existing material; ambiguous cases surfaced instead.
• Invalidation or expiry: Next audit reassesses; repair metadata updated; no atomic rollback specified.
• Activation or effect: Requested autonomous content repair within semantic constraints; not a losslessness invariant.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

A fix is safe and deterministic when it loses no meaning, invents no value, and resolves no genuine ambiguity. --- plugins/knowledge_management/agents/auto_shaper_wiki.md

EPI-RTE-wiki-archive

Label: Wiki active withdrawal

Part of: RT-RTE-wiki-consume

Separate material function within that container; no supersession needed.

Foreground model moves fully superseded or out-of-scope page to _archive preserving its relative path, removes index entry and converts inbound links to plain archived text. Target is active navigation membership, not refutation of source truth. Model identifies supersession under user request/current scope; true conflict resolution remains human-owned. Consumer is later index/link-based query and linter page walk; force is omission from active navigation and default page walk, horizon until restoration/reorganization. Archive retains the body and does not establish all recursive host searches exclude it: the query search is across all Markdown. Thus active withdrawal is not total access revocation. Lint catches remaining broken inbound paths; no automatic multi-file rollback.

Endpoints/progression: selected inputs → wiki active withdrawal → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-wiki-pages and the described input/result/consumer chain. Evidence: SRC-1 plugins/knowledge_management/skills/wiki/SKILL.md.

• Immediate return: Archive path, navigation/link changes and log/lint report.
• Later read-back: Archived body remains on disk; changed navigation affects later ordinary lookup.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Fully superseded content or changed domain scope.
• Invalidation or expiry: No expiry or automatic restore rule found in inspected archive block.
• Activation or effect: Requested reduced active selection, not epistemic disproof or complete retrieval prohibition.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

Update inbound links: replace with plain text + "(archived)". --- plugins/knowledge_management/skills/wiki/SKILL.md

EPI-RTE-text-check

Label: Source fidelity self-check

Distinct identity: no counterpart after comparison with supplied runtime route declarations. This downstream/function-specific operation was not separately declared there.

Model compares delivered text to its own input ledger: every load-bearing element must retain strength/scope; missing items are restored before return. SPR instead checks paragraph/bullet mapping and emits reconstructability rating; executive_summary checks required format/length/coverage and self-rates. Targets are textual fidelity/coverage and requested audience readability, not independent source truth. User supplies input/reader, model judges; ambiguities return to author. Same-model self-check and a numeric self-rating are not independent semantic or reconstruction evidence. This is distinct from text production in EPI-RTE-text-reshape; both are standalone skills with no counterpart in supplied runtime routes.

Endpoints/progression: selected inputs → source fidelity self-check → named result/consumer described above. Owner: foreground/helper model following skill prose; host owns actual execution. Context/state/action effects: see EPI-OBJ-text-output and the described input/result/consumer chain. Evidence: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md.

• Immediate return: Restored delivered text, preservation note or self-rating; review mode reports draft findings.
• Later read-back: inapplicable — no dedicated retained evaluation store or later consumption rule is specified.
• Delegated visibility: uninspected — host context assembly and helper input execution were not observed.
• Selection predicate: Before returning rewrite/write text; review uses draft ledger as yardstick.
• Invalidation or expiry: Author correction can prompt another pass; no automatic expiry/rollback.
• Activation or effect: Requested revise-before-return feedback; actual successful preservation unobserved.
• Evidence limits: Inspected doctrine/design only; model compliance, produced instances and successful checks not observed.
• design conclusion status: claimed
• operation conclusion status: uninspected

confirm each item is present with its strength and scope unchanged, restoring anything missing before returning it. --- plugins/ai_editorial/skills/language_humanizer/SKILL.md

Supplementary evidence for the differentiated consumers:

When every body item, acceptance check, required test, and the design-extended signal check is confirmed and the current task status is implemented, stamp status: audited and bump updated. --- plugins/ai_dev/skills/task_audit/SKILL.md

body = text[end + 5:] --- plugins/knowledge_management/skills/wiki/scripts/compute_sha256.py

actual = hashlib.sha256(body.encode("utf-8")).hexdigest() --- plugins/knowledge_management/skills/wiki/scripts/compute_sha256.py

LLMs hold latent knowledge, reasoning, planning, and world-modeling capacities that selectively activate through priming, the specific token sequences that evoke internal neural states. --- plugins/knowledge_management/skills/spr/SKILL.md

Use only INPUT content; introduce no additions or assumptions. --- plugins/knowledge_management/skills/executive_summary/SKILL.md

This line stands in place of asking permission before filing: decide on the trigger, report the call, and let the user correct it in the next turn, --- plugins/knowledge_management/skills/wiki/SKILL.md

if any(a.matches(issue, rel) for a in acceptances): acknowledged.append(issue) else: live.append(issue) --- plugins/knowledge_management/skills/wiki/scripts/lint.py

if quiet: issues = [i for i in issues if i.severity != SEV_INFO] ack = [] --- plugins/knowledge_management/skills/wiki/scripts/lint.py

Set confidence: medium or low on single-source, opinion-heavy, or fast-moving pages. Reserve high for pages with multi-source support. --- plugins/knowledge_management/agents/auto_shaper_wiki.md

when it is audited, treat the codebase-verification gate as already satisfied and proceed to close-out --- plugins/ai_dev/skills/task_finish/SKILL.md

integrity_ok = not isinstance(integrity, dict) or bool(integrity.get("passed", True)) --- tests/wiki/layer2/grade.py

worker_ok = worker_rc == 0 and bool(response_text.strip()) --- tests/wiki/layer2/grade.py

ok = target.is_file() and target.stat().st_size > 0 --- tests/wiki/layer2/grade.py

pass_rates = [p[1]["grading"].get("pass_rate", 0.0) for p in passes] --- tests/wiki/layer2/aggregate.py

if regressions: print(f"\n!! REGRESSIONS DETECTED ({len(regressions)}):") for r in regressions: print(f" - {r}") return 1 return 0 --- tests/wiki/layer2/aggregate.py

rc = subprocess.run(cmd).returncode --- tests/wiki/layer2/run.py

Claims

EPI-CLM-spr-recoverability

Label: SPR semantic recoverability

Claimed operation: source-bounded priming lets an unexposed receiving LLM reconstruct original material with maximal semantic recoverability. Evidence: SRC-1 plugins/knowledge_management/skills/spr/SKILL.md, doctrine/design. Distinct from RT-CLM-persistent-context: a receiver-reconstruction claim, not persistent filesystem context. See EPI-RTE-text-reshape; self-rating is not a receiver test or independent truth oracle.

Maximize semantic recoverability by encoding the deepest structure and nuance of the input with minimal tokens. --- plugins/knowledge_management/skills/spr/SKILL.md

Stay strictly within the source's scope; introduce no content that is not derivable from the input. --- plugins/knowledge_management/skills/spr/SKILL.md

EPI-CLM-shaper-safety

Label: Unattended safe wiki remediation

Claimed operation: lossless, value-preserving repairs and surfaced ambiguity make unattended maintenance safe. Evidence: SRC-1 plugins/knowledge_management/agents/auto_shaper_wiki.md, doctrine/design. Distinct from supplied context/callability claims: concerns semantic repair safety. Routes EPI-RTE-wiki-critic, EPI-RTE-wiki-dispute. Mechanical lint cannot independently establish absence of invented meaning.

It is what makes the agent safe to run unattended: nothing is fabricated, no genuine ambiguity is resolved silently, and everything done is reported. --- plugins/knowledge_management/agents/auto_shaper_wiki.md

EPI-CLM-editor-fidelity

Label: Humanizer content preservation

Claimed operation: every load-bearing input element reaches clear delivered text unchanged in force/scope. Evidence: SRC-1 plugins/ai_editorial/skills/language_humanizer/SKILL.md, doctrine/design. No supplied counterpart: standalone editorial fidelity, not retained wiki context. See EPI-RTE-text-reshape.

Where a mode returns text, walk the ledger item by item against the text about to be returned and confirm each item is present with its strength and scope unchanged, restoring anything missing before returning it. --- plugins/ai_editorial/skills/language_humanizer/SKILL.md

Evidenced absences

none declared in this member

Behavioral-authority paths

none declared in this member; differentiated paths are recorded in the route declarations and supplied RT-BAP-wiki-guidance, RT-BAP-task-guidance, RT-BAP-skill-guidance.