Agentic system analysis run state

Type: types/type-spec.md

Authoring Instructions

Use this type only for kb/reports/state/agentic-system-analysis/<run-id>/run-state.md. The owning workflow is kb/instructions/analyse-agentic-system/SKILL.md.

This record proves only what later consumers need:

  • which run produced the outputs;
  • which frozen source boundary it used;
  • which exact result and published review bytes completed the run; and
  • which typed memory specialist report the exact result integrated.

It is not a recovery log. A run is running, complete, or failed. Do not resume a failed run or preserve phase, packet, correction, validation-receipt, or retry state. Start another run with a new run ID. Temporary candidate files inside the run directory are disposable and never appear in this record. Keep the destination inspection's expected incumbent digest (or absent) in ## Run prose. Publication requires it again at prepare and publish. Recovery copies incumbent-review.md and incumbent-result.md are retained with the new run when a review is replaced; they are not disposable candidate files.

The exact result always lives at kb/reports/state/agentic-system-analysis/<run-id>/result.md. A substantive complete result also publishes one generated review under kb/agentic-systems/reviews/. A blocked or out-of-scope result has no generated review. Publication also retains the exact result bytes at kb/reports/retained/agentic-system-analysis/<run-id>/result.md. Its identity is derived from the run ID and the existing result.sha256; no duplicate output mapping is needed. Completion verification checks this copy and the public review's analysis-result path and analysis-result-sha256. Durable comparison readers follow those public fields without requiring ignored run state or a local source checkout. Every substantive complete analysis requires the typed memory-report.md and frozen memory-input.md in its run directory. The exact result names the report and its SHA-256 in Run identity. Completion checks the report's run, source, reviewed boundary, complete status, and input hash, and validates its type and source anchors. These checks establish identity and structure; they do not certify the specialist's semantic judgments.

source is either a Git commit or an immutable capture. A Git source records the stable repository identity, full commit ID, and absolute checkout path. A capture records its stable identity, version or capture label, absolute file path, and SHA-256. The validator checks the commit or capture while the run state exists.

For a Git source, replace source: null with this mapping, substituting the repository identity, full commit and absolute checkout path:

source:
  kind: git
  identity: https://github.com/owner/repository
  revision: "0123456789abcdef0123456789abcdef01234567"
  path: /absolute/path/to/checkout
  sha256: null

Use path, not root. Git sources require sha256: null; immutable captures use their content digest instead. Validate the running state again immediately after setting the source, before source analysis or delegation.

Each output mapping contains a normalized repository-relative kb/ path and the SHA-256 of its current bytes. Validate candidate bytes as their intended destination before publication. The run-state validator rechecks byte, workflow, and specialist handoff identity; it does not retain a validation receipt.

A publication candidate is validated before it replaces a same-source review. A correctable failure before publication leaves the incumbent unchanged and the run running. Mark a run failed only when abandoning it or when a publication failure leaves public state uncertain. Git history is the history of successfully published tracked reviews; this workflow does not stage or commit.

Template

---
type: types/agentic-system-analysis-run-state.md
description: "Minimal completion state for AAS-YYYY-MM-DD-system-slug-nn"
run-id: AAS-YYYY-MM-DD-system-slug-nn
system: "Source-native system name"
run-status: running
result-disposition: null
source: null
result: null
generated-review: null
failure: null
---

# Agentic-system analysis run — AAS-YYYY-MM-DD-system-slug-nn

## Run

<Target and frozen source boundary once known.>

## Outcome

<Current status, completed outputs, or concise failure reason.>