sign-lang boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

This run analyses SIGN as an agent-oriented knowledge contract: its notation, authored knowledge documents, validation and compilation gates, and source-coverage and change signals. It is in scope as a knowledge/context-engineering system, not as an enclosing agent runtime. The intended consumer is explicit in SRC-1, README.md:

SIGN is the contract layer between a governed knowledge graph and the agents that consume it. It expresses structure, relationships, inference rules, constraints, clusters, and provenance in a form agents can reason over — not just retrieve from. --- README.md

The boundary is complete artifact, partial loop. The frozen repository provides the language and deterministic preparation/maintenance toolchain, but not an implemented enclosing model-consumption loop. The README's injection and get_canon_doc(id) snippets describe integration responsibilities; they are not a shipped provider client or retrieval service. The complete tracked tree was inspected, the CLI was traced into its modules, and the editor extension manifest was checked for additional execution entry points. The local .agent/sign/ corpus is included as shipped knowledge and authoring rules, not silently excluded as a library caller. These checks establish the artifact boundary, not an end-to-end operational agent system (SRC-1, README.md, validator/bin/sign.js, editor-support/vscode/package.json, .agent/sign/canon.source.json).

The declared delivery split is:

| Layer 1 — Index | Injected into every agent system prompt. ~400 tokens for a full corpus. Document identity, status, graph edges, one summary line per document. Orientation without rule text. | | Layer 2 — Full document | Retrieved on demand via get_canon_doc(id). Agents only pay for documents they actually need. | --- README.md

Included executable responsibilities are source discovery and parsing; structural and bundle-level rules; compilation to persisted index, full bundle, manifest and checksum; repository-file coverage; Git-based drift; document-change classification; and the CLI's external-import delegation boundary. The compiler reads .sign sources, runs rules with strict link checking, selects active document identities, builds the index, passes through raw files contributing active documents for the full bundle, and writes four artifacts. This is not evidence that arbitrary Markdown in the README quick start is compiled by the inspected implementation (SRC-1, validator/src/parse.js, validator/src/rules.js, validator/src/validate.js, validator/src/compile.js). Its persisted artifact interface is directly visible:

fs.writeFileSync(path.join(outputPath, 'canon-index.sign'), finalIndex, 'utf8'); fs.writeFileSync(path.join(outputPath, 'canon-bundle.sign'), bundleContent, 'utf8'); fs.writeFileSync(path.join(outputPath, 'canon-manifest.json'), JSON.stringify(manifest, null, 2), 'utf8'); fs.writeFileSync(path.join(outputPath, 'canon-bundle.sign.sha256'), sha256 + '\n', 'utf8'); --- validator/src/compile.js

Maintenance signals remain distinct from a review workflow. The drift module itself states the split (SRC-1, validator/src/drift.js):

// (no platform dependency); the platform layer turns these signals into review workflow. --- validator/src/drift.js

Inspected coverage and remaining scope

This is not a whole-system coverage claim. All tracked paths at the pin remain available for later inspection under SRC-1; the initial reads below do not restrict that source to an allowlist.

External dependencies are Node.js (package engine requirement >=18), local filesystem access, Git for historical comparisons, and a separately discovered @sign-tools/sign-ingest process for sign import. The CLI's discovery/spawn/error behavior is included; that other package's internals and outputs are not frozen here. Likewise the external sign-draft skill copy and ch-canon-tools/platform work mentioned in the inspected documentation are not acquired sources. Their exclusion prevents conclusions about ingestion, model drafting, human landing, platform resolution or review orchestration beyond this repository's declared interfaces (SRC-1, validator/package.json, validator/bin/sign.js, spec/skill-reference.md, proposals/2026-07-codebase-canon-reinforcement.md).

The excluded enclosing application's model selection, prompt assembly in production, retrieval implementation, enforcement of semantic constraints, durable agent memory and feedback loop must not be attributed to this toolchain. No execution traces, provider calls, benchmarks or causal experiments were supplied or performed. README production-validation and efficiency statements remain attributed claims only. The frozen evidence is the Git repository at the full commit above, verified against checkout HEAD with a clean status; applicability ends at the run's 2026-11-09 cutoff. Overall evidence is code-grounded for the inspected toolchain, with doctrine/design evidence for intended agent use.

Source register