AgentStack
SKILL verified MIT Self-run

Reconcile

skill-aigengame-godot-agent-reconcile · by aigengame

Check and repair consistency between GitHub issues and domain docs (CONTEXT.md, docs/adr/) after a requirement or technical decision changes — mechanical cross-reference integrity plus conversation-driven semantic staleness. Use when a decision or requirement was changed or discussed in the current session, when issues/docs may have drifted out of sync, when the user says "reconcile" / "sync issu…

No reviews yet
0 installs
16 views
0.0% view→install

Install

$ agentstack add skill-aigengame-godot-agent-reconcile

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Reconcile? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Reconcile

Keep the cross-referenced web of issues + domain docs (CONTEXT.md, docs/adr/) correct and consistent so the development feedback loop stays trustworthy. A decision discussed in a session must not silently drift out of sync with the issues and docs that record it.

Read first: docs/agents/domain.md (glossary + ADR-conflict rules), docs/agents/issue-tracker.md (gh conventions), docs/agents/triage-labels.md. See [REFERENCE.md](REFERENCE.md) for the reference-graph edge taxonomy and the full check catalog.

Triggers

  • Manual: invoked as /reconcile, or when the user asks to sync issues and docs.
  • Automatic: a Claude Code Stop hook (and optionally a git pre-commit hook) may auto-invoke

this skill — see "Hook setup" below. Hook wiring is HITL; never add it without user approval.

Safety posture (non-negotiable)

Report first, mutate only after confirmation. Always present findings and proposed patches, then wait for the user to confirm before editing any doc file or GitHub issue. Never run a mutating gh issue edit / gh issue comment or write to a doc without explicit approval.

Workflow

gather change set → build reference graph → check (mechanical + semantic) → report → confirm → apply + propagate

1. Gather the change set

The semantic layer is driven by the current conversation context. Identify the decision / requirement changes discussed in-session: which terms were renamed, which decisions reversed or refined, which scope/Phase boundaries moved. Write them down as an explicit change list. If the session contains no such change, run the mechanical layer only and say so.

If the user states a change explicitly, use that verbatim as the change set.

2. Build the reference graph

Load the artifacts and extract their cross-references (see [REFERENCE.md](REFERENCE.md) for edge types and how to read each source):

  • Issues + PRDs — gh issue list/view (Parent, Blocked by, ADR-NNNN, #NN, domain terms)
  • ADRs — docs/adr/*.md (ADR-NNNN cross-refs, status: front-matter, domain terms)
  • CONTEXT.md glossary — canonical terms and their _Avoid_ synonyms; ADR-NNNN refs

3. Check — two layers

Mechanical (always runs): dangling references (ADR-9999, #999, missing file paths), orphans, and glossary term drift (use of an _Avoid_ synonym instead of the canonical term).

Semantic (runs when a change set exists): for each change, find the issues / ADRs / CONTEXT.md entries it makes stale — a change that contradicts an ADR's Decision, a renamed term, a moved Phase boundary, an issue describing superseded behavior. Report each with the reasoning that links the change to the affected artifact.

4. Report

Present a single consolidated report: every finding with its evidence (file/issue + line/quote) and a concrete proposed patch (a diff for docs, the exact gh command for issues). Group by artifact. Flag any finding that contradicts an accepted ADR per domain.md's flag rule.

5. Confirm (HITL)

Ask the user which proposed patches to apply. Do not proceed on un-confirmed items.

6. Apply + propagate

For confirmed fixes, apply them, then propagate consistency across the graph:

  • Edit doc files; for issues use gh issue edit / gh issue comment per issue-tracker.md.
  • When an ADR is superseded, add a superseded-by link both ways and update its status:.
  • Sync the CONTEXT.md glossary when a term changes (definition, _Avoid_ list, ADR refs).
  • Comment affected issues so the change is traceable in the tracker.
  • Re-run the mechanical check on touched artifacts to confirm no new dangling references.

Hook setup (optional, HITL)

To auto-invoke on a trigger, propose (do not silently add) a hook in settings.json — a Stop hook for session-end reconciliation, or a git pre-commit hook. Review timing and behavior with the user before writing to settings.json. See [REFERENCE.md](REFERENCE.md) for a sample.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.