nhhthong
UserA project-memory kit for .claude/: layout, ledgers, and the loop that keeps them true.
Categories
Indexed Skills (10)
update
After a spec file changes, work out what the change means for code already written — which task/decision docs it invalidates, whether it is safe to implement yet — record each delta in .claude/clio/debt.jsonl, and move the requirements.md row marker when the decision itself changed. Run when a requirement, contract or ticket changes.
context
Load the spec, prior task docs and open debt for an area before working in it. Use before any non-trivial task, and when the user asks "where are we", "what's next", "why is X like this" or "any history on this".
ask
Explain Clio — what each clio:* command does, when to run it, which files it reads and writes, and how the three ledgers work. Use when the user asks "what does /clio:memo do", "how does Clio work", "which clio command should I run", or anything about the Clio plugin.
ingest
The only writer of the spec layer. First run — turn a requirement document (contract, PRD, spec.md, ticket export) into one .claude/clio/docs/specs/memory/*.md per area, the requirements.md row table, and the stack as memory/infra.md (row 0, ADR), marking every undecided point ⚠️. Every later run — a new source, or no argument to sweep spec files edited by hand — also works out what the change invalidates in code already built, files spec-delta records, moves row markers and names the areas to re-plan. Run when requirements arrive or change.
audit
Check the three layers against each other — requirement rows, plan tasks and the ledgers — and name every place they have drifted apart. A spec that moved after its area was planned is the common one, and nothing else notices it. Read-only by default; --fix re-plans the areas that drifted. Run when a spec changed, or any time you want to know whether the plans still match what was decided.
migrate
Migrate .claude/ onto the layout this plugin version expects — move task docs into their feature directory, give pre-3.0 docs an id, append full records where an old one is short, strip leftover HTML comments. Dry-run by default; --fix asks before it writes anything. Run after upgrading the Clio plugin, or when validate.sh keeps naming the same old records.
memo
Record a finished piece of work — update (or create) its sub-task doc under .claude/clio/docs/tasks/<feature>/, append the run to .claude/clio/database/index.jsonl, log what is still owed to .claude/clio/database/debt.jsonl, tick the plan task once /clio:test's gate passes, and propose a path-scoped .claude/rules/ entry or an ADR when warranted. Works on committed or uncommitted work, and backfills the commit hash later. Run after finishing a feature, fix or refactor.
setup
Set up Clio in the current project — create .claude/clio/docs/ (specs, plans, tests, tasks, decisions) and .claude/clio/database/ (the ledgers), asking the user only what nothing on disk can tell. Never touches CLAUDE.md, CONTEXT.md, .claude/rules/ or the stack. Run once per repository, before any other clio:* skill.
plan
Break a spec area into the smallest independently testable tasks — one observable behaviour each, researched against the code and the library docs, with the test levels it needs (unit, api, security, concurrency…) — written to .claude/clio/docs/plans/<area>.md in dependency order. `infra` goes first: it turns the stack /clio:ingest decided (memory/infra.md), or the stack the repo already has, into toolchain and scaffold tasks and writes .claude/rules/<stack>.md. Never chooses a stack. Run `infra` after /clio:ingest, an area after that, and again after /clio:ingest moves a row.
debt
Show what is still owed — open records in .claude/clio/debt.jsonl split into actionable-now vs blocked, optionally filtered by domain, requirement row number or keyword. Read-only.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.