agentsclimarketplace

Recall prior work

Skill arniesaha/ctxops/skills/recall-prior-work

Agent skills (open standard) for the bookends of the agent loop: context assembly, decision review, and work-shaping outputs. Works with Claude Code and any Agent Skills harness. Ships its own adoption and value measurement.

Install
npx -y skills add arniesaha/ctxops --skill recall-prior-work

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

WHEN the user references past sessions, prior decisions, or work outside current context: search persistent memory FIRST before answering. Triggers: 'did we already X?', 'how did we solve Y last time?', 'what was decided about Z?', 'last time we tried this…'. Requires a persistent-memory tool (e.g. claude-mem); degrades to git history + docs without one. OUTPUT: found / decided / tried-but-unresolved / stale-risk, with dates. PAIR: feed results into ctx-pack or status-update. NOT FOR: current-session recall (it's in context), repo-state questions (read the repo).

SKILL.md

2.4 KB, as published. Nobody here has run it

recall-prior-work

Memory-only lookup, distinct from ctx-pack (which also pulls memory but blends it with files/diffs) and project-briefing (which reads the live repo).

Workflow

Three layers, always in order. Never fetch full details before filtering. Tool names below are claude-mem's; map them to whatever memory tool is installed. If none is, say so and fall back to git history, status docs, and ADRs. Do not fabricate continuity.

  1. Search - memory search with a focused query, project filter, and a date window if known. Read titles only.
  2. Timeline - if a single hit looks load-bearing, pull timeline around it for surrounding context.
  3. Fetch - get_observations on the small set of IDs that actually matter. Batch the call.

Then:

  • Distinguish what was decided from what was tried. Decisions stick; experiments may not.
  • Note the date of each fact. Memory is a snapshot, not ground truth.
  • Verify load-bearing claims (file paths, flag names, ticket IDs) against the current repo before recommending action.

Output Shape

  • Found - 1-line per relevant memory with date and ID.
  • Decided - what was settled, and when.
  • Tried but unresolved - so we do not re-litigate or re-attempt blindly.
  • Stale risk - facts whose freshness matters; what to re-verify.
  • Next step - either an answer or the verification command.

Rules

  • Filter before fetching. Titles first, full text only for finalists.
  • If memory and current repo disagree, trust the repo and update memory.
  • Do not paste long memory excerpts into the response; cite IDs and summarize.
  • If no relevant memory exists, say so explicitly. Do not fabricate continuity.
  • Respect project scope: pass the project filter so memory from unrelated repos does not leak in.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.