| name | speckit-archive-cleanup |
| description | Archive a merged SpecKit spec and clean active workflow residue after the implementation PR has merged. Use after confirming merge provenance, when the user asks for post-merge SpecKit archive hygiene, cleanup hygiene, or removal of completed specs from active specs. |
| argument-hint | SPEC-ID and optional merged PR URL or number |
| user-invocable | true |
| allowed-tools | Read Edit Write Grep Agent |
| license | MIT |
SpecKit Archive Cleanup
Inputs
Accept a SPEC-ID such as DOC-007 or PRSG-014, an active spec directory, a
workflow file path, or a merged PR URL/number. If more than one is provided,
cross-check that they all point to the same completed work.
Required facts before editing:
- merged PR number, merge timestamp, merge commit, and PR title
- active spec directory under
specs/
- workflow file under
docs/ai/specs/.process/, if present
- current roadmap and traceability files affected by the spec family
- installed archive extension contract, if
.specify/extensions/archive/ exists
Ground Truth Checks
Start from live repository truth:
- Inspect
git status --short --branch.
- Confirm the current branch is a cleanup branch based on the current mainline,
or create one before editing.
- Confirm the PR is merged with GitHub tooling or the best available local
merge evidence.
- Read the existing newest archive reports in
.specify/memory/archive-reports/
to match local conventions.
- Check whether
.specify/feature.json exists. If it is absent, do not create
it. If it exists and points at the completed spec, remove or rewrite it only
according to repository convention.
- List active specs with
find specs -mindepth 1 -maxdepth 4 -print and
identify the exact folder that belongs to the merged spec.
Do not remove any active spec folder until merge provenance and recovery
commands are recorded. Do not remove process files under
docs/ai/specs/.process/ unless repository history shows that process evidence
is intentionally deleted for completed specs. In this repository, process files
are preserved as historical evidence.
Archive Procedure
Read the archive extension command contract before making archive edits when it
is present. Treat it as the local policy for source directories, memory files,
cleanup eligibility, and extension hooks.
Then update the project state in this order:
- Add an archive report under
.specify/memory/archive-reports/ named with the
current date and SPEC-ID. Include PR URL, merge commit, merged-at timestamp,
source spec path, workflow file, canonical shipped artifacts, cleanup branch,
cleanup command, verification commands, and exact recovery commands using
git show or git checkout against the merge commit.
- Append concise records to
.specify/memory/spec.md,
.specify/memory/plan.md, and .specify/memory/changelog.md. These records
should summarize what shipped, where canonical artifacts live now, why the
active spec folder can be removed, and where the detailed archive report is.
- Update roadmap, traceability, agent context (AGENTS/CLAUDE/GEMINI), or MOC
files ONLY to remove or correct references that still describe the merged
spec as pending, in progress, or blocking downstream work. Never append
per-spec history entries (archive notes, Active Technologies bullets, or
Recent Changes bullets) to agent context files — the archive report and
.specify/memory/ records are the system of record for history, and agent
context files must stay small (Codex reads AGENTS.md under a 32 KiB budget).
- Update
docs/ai/specs/.process/autopilot-state.json only if it exists and
still points at the completed spec. The status should become an archived or
completed archive state, with the cleanup applied and post-merge archive
phase completed.
- Remove the completed active spec directory under
specs/. Keep specs/.gitkeep.
- Regenerate the active spec index with the repository's existing generator,
then run its
--check mode.
Prefer local helper scripts over hand-maintaining generated files. If the repo
has docs-site generated reference pages or generated plugin payloads affected by
the cleanup, run the relevant generators and include those generated changes.
Safe Parallelism
These parts are safe to do in parallel:
- read-only discovery such as
git status, gh pr view, find specs, and
reading roadmap, memory, and workflow files
- inspecting multiple archive reports
- running independent read-only searches for stale SPEC-ID mentions
These parts must be serialized:
- edits to roadmap, traceability, memory, MOC, and autopilot-state files
- active spec directory removal
- generated index and generated docs updates
- staging, committing, pushing, and PR creation
The serialized files all represent one shared project state. Parallel edits
make it easy to leave contradictory status such as "archived" in memory but
"in progress" in a roadmap.
Verification
Run the smallest checks that prove the cleanup, then the standard project
checks if plugin or generated payload files changed. Typical checks:
- active spec listing shows only expected active specs and
specs/.gitkeep
python3 -m json.tool docs/ai/specs/.process/autopilot-state.json- SpecKit index generation and
--check
- docs-site reference generation/checks when reference pages changed
- payload builder and payload parity checks when plugin source changed
git diff --check- repository structural validation suite
If a check cannot run, report the exact command and the reason. Do not claim the
archive is fully verified when generated files or structural checks are stale.
Git And PR Titles
For archive-only cleanup commits and PRs, use a lower-case Conventional Commit
scope derived from the completed spec ID. For example, archive cleanup for
CAR-001 should use docs(car-001): archive post-merge state, not
docs(CAR-001): archive post-merge state. The repository PR title gate checks
the final PR title, so apply the same lower-case scope to gh pr create or
gh pr edit --title.
Final Report
Report:
- the merged PR and merge commit used as provenance
- the active spec folder removed
- archive report path
- roadmap or traceability status changes
- generated files refreshed
- verification commands and results
- remaining risks, especially skipped browser UAT or skipped CI checks
Keep the report short and make the next action explicit, usually review the
cleanup PR or merge it after CI passes.
