Move intent/<slug>/ to intent/archive/YYYY-MM-DD-<slug>/. Nothing is deleted and nothing is rewritten: the folder keeps the intent, the spec, the plan, and the report exactly as they were.
An archive is decision history. Six months on, the question is why this was built this way, and the answer is the folder.
Before you start
Invoking this workflow normally authorizes the archive commit, subject to explicit user restrictions and host policy. Before moving anything, record repository, branch, HEAD, staged paths, and working-tree/untracked paths. Preserve unrelated work; if the index contains someone else's staged changes or ownership overlaps, use a clean isolated worktree when feasible or stop the commit. Never reset, stash, or absorb those changes. Use an ordinary filesystem directory move with the host's available tools; Git recognizes the rename from content. Keep every artifact byte intact and require an absent destination. Stage only the move. Follow Conventional Commits 1.0.0: chore(scope): imperative summary (scope optional), a blank line, and one sentence on why. Do not push unless asked.
intent/archive/ is the archive, not a slug. Skip it when you list changes.
Steps
-
Resolve the slug. In an ordinary run, say which one you picked; if several are candidates, ask. In an explicitly authorized autonomous queue, use its selected slug without another approval request.
-
Resolve this skill's installed directory, then run
python3 "<this-skill-dir>/scripts/validator.py" archive-check <slug>from the product repo root. Use an available Python 3.8+ interpreter if named differently; resource paths are skill-relative, not product-repo-relative. This local copy uses the same contract assdlc-continue. It checks grammar, all done statuses, plan boxes, approval digests and upstream bindings, substantive report evidence, valid independent isolation, reviewed HEAD freshness, and dirty/untracked paths. Read its reason on failure and follow the owning skill; never waive a failed check. A blocked verification needs recovery, while a failed verification needs repair and a fresh review. -
Read the report and compare its claimed observations with the cited evidence. Mechanical validity cannot prove that the verifier actually exercised a flow. If a cited required observation is absent or contradicted, withhold archive and request fresh independent verification.
-
Build the target name. Today's date as
YYYY-MM-DD-<slug>, unless the slug already starts with aYYYY-MM-DD-prefix, in which case use it as is. Never stack a second date. -
If
intent/archive/<target>already exists, stop. Do not overwrite or merge. Tell the user, and let them rename the existing archive or pick another date. -
Create the archive parent if needed and move the whole folder with the host's filesystem tools, without overwriting an existing destination. For a POSIX shell:
mkdir -p intent/archive mv "intent/<slug>" "intent/archive/<target>" -
Confirm the destination path and tasks tally. Then commit the move. Archive closes the implementation record; it does not assert deployment or production success.
After the verified archive move, make a bounded knowledge check only if the project or environment already configures a knowledge bundle or maintenance workflow. Use that workflow to update a relevant existing durable note when this work supersedes it; otherwise skip. Do not create a note just to record the session, scan the whole bundle or unrelated knowledge, or bootstrap a missing bundle. This adds no delivery gate. Keep knowledge edits separate from the reviewed candidate and archived artifacts; the archived folder remains byte-for-byte intact and the archive commit stages only the move.