Reconcile
SkillMediaCheck and repair consistency among tracked work, requirements, design records, glossaries, and other authoritative project artifacts after a requirement, decision, scope, or term changes. Use when artifacts may have drifted out of sync, when the user asks to reconcile or check cross-references, or after the current session changes an authoritative fact.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Reconcile skill
What this skill tells your AI
The instructions your AI receives, as published by aigengame/godot-agent in .agents/skills/reconcile/SKILL.md and read by ahel’s review.
Keep related project artifacts consistent after an authoritative fact changes. Reconcile both structural references and statements whose meaning became stale.
Do not assume a fixed tracker, document name, directory layout, identifier format, or decision-record schema. Read the target repository's guidance first. Use its context routing, artifact ownership, terminology sources, and editing rules as the authority. See REFERENCE.md for the reference-graph taxonomy and check catalog.
Triggers
- Manual invocation: run when the user asks to reconcile artifacts or check cross-references.
- Optional reminder: a session-end or version-control hook may remind the user to run this skill. A reminder does not run reconciliation. Call a hook automatic only when the host has a documented, verified mechanism that invokes the skill. Do not add or change hook configuration without the user's approval.
Safety posture
Report first. Mutate only after confirmation. Present findings and proposed changes before editing local artifacts or remote tracker records. Apply only the changes that the user confirms.
Workflow
discover authorities -> gather change set -> build reference graph -> check -> report -> confirm -> apply and propagate
1. Discover authorities and scope
Read the repository guidance that defines artifact locations, context boundaries, terminology, tracker conventions, and decision-record rules. Follow nested guidance for the target area.
Determine whether the repository has one context or several. For a scoped change, use the owning context and every routed context that consumes the changed fact. For a repository-wide check, enumerate all declared contexts. If no routing authority exists, discover the relevant artifacts from repository guidance and the artifact links themselves.
Record which artifact owns each affected fact. Treat other occurrences as references or derived statements unless the repository declares shared ownership.
2. Gather the change set
Use the current conversation and authoritative artifacts to list the requirements, decisions, terms, or boundaries that changed. Preserve an explicit user statement exactly when it defines the change. If there is no semantic change set, run only the mechanical checks and report that limit.
3. Build the reference graph
Load the artifacts in scope and extract their declared references. Typical nodes include tracked work, requirements, decision records, specifications, glossaries, tests, and implementation docs. Use repository-native identifiers and relationships. Do not infer an edge from a shared keyword alone.
4. Check two layers
Mechanical checks always run. Check facts that can be decided without interpreting intent:
- a declared reference target exists;
- an artifact does not use a synonym that the terminology authority explicitly prohibits;
- a reciprocal pointer exists when the repository schema explicitly requires one.
Semantic checks run when interpretation is required. For each change, find statements that may now be stale. Check changed decisions, renamed or newly introduced terms, moved scope boundaries, partly or fully superseded behavior, tracker state, and unreferenced artifacts. Use evidence to explain why each candidate is or is not stale. Do not classify tracker state or glossary coverage as a defect without applying the repository's meaning and rules.
5. Report
Present one report grouped by artifact. For every finding, provide the location, evidence, impact, and a concrete proposed change. Use a patch for local text. For a remote record, describe the exact edit using the target repository's tracker and tool conventions. Flag any proposal that conflicts with an accepted authority; do not silently override it.
6. Confirm
Ask which proposed changes to apply. Do not proceed with unconfirmed items.
7. Apply and propagate
Apply confirmed changes, then update every affected reference or derived statement:
- Use the repository-native tool and editing rules for each artifact.
- For complete supersession, update the old and new decision records as the repository schema requires.
- For partial supersession or refinement, keep the still-valid decision active. State the affected scope in both records when the repository requires bidirectional traceability.
- Do not invent frontmatter fields, statuses, or link directions. If the repository has no supersession convention, propose one separately instead of adding an implicit schema.
- Update the terminology authority and affected uses when an established term changes.
- Add tracker traceability only when repository guidance requires it.
- Re-run the applicable mechanical checks on every touched artifact.
Reminder setup (optional)
Propose a reminder only when it helps the repository workflow. Review its timing, visibility, and host-specific behavior with the user before changing configuration. The reminder must remain non-blocking because reconciliation can require confirmation. See REFERENCE.md for the reminder contract.
Signals
- GitHub stars
- 38
- Forks
- 6
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
reconcile- Source
- github.com/aigengame/godot-agent