agentsclimarketplace

Recalling recipes

Skill ljf06853/keepfire/skills/recalling-recipes

Use when the user wants to reuse a past prompt pattern — "use my usual PR review style", "用我以前那种方式", /use — AND before starting any substantial implement/debug/review/refactor/test task: check whether the user's recipe library already holds a proven approach before inventing your own.From its SKILL.md

Install
npx -y skills add ljf06853/keepfire --skill recalling-recipes

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

  • 1 stars1 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.

SKILL.md

2.6 KB, 589 tokens by cl100k_base, as published. Nobody here has run it

Recalling Recipes

The user maintains a library of proven prompt recipes under ~/.keepfire/cards/. Reusing a proven recipe beats inventing an approach from scratch — but only if it is adapted to the current task.

<HARD-GATE> NEVER paste an old prompt blindly. Always adapt: fill placeholders from the current task and repository, drop stale file names and dead constraints, keep the recipe's structure, constraints, and quality signals. </HARD-GATE>

Two entry modes

Explicit (user asked to reuse a style): run the full checklist below.

Ambient (you are about to start a substantial coding task): do a quick search first. If a strong match exists, offer it in one line — 🔥 你有一条相关配方「<title>」,要用吗? — and continue only on yes. If nothing matches well, proceed silently; never nag.

Checklist

  1. Search the library:
keepfire search "task keywords" --top 5

No CLI? Grep ~/.keepfire/cards/*.md titles, trigger_phrases, and tags yourself. Empty library → tell the user how to start: say "keep this prompt" after a good one.

  1. Present the top 1–3 candidates with id, intent, and why they matched.

  2. Select per use_mode in ~/.keepfire/config.json:

    • always_ask — let the user pick
    • ask_if_low_confidence (default) — auto-apply only a strong match, otherwise ask
    • auto — apply the best match
  3. Render the adapted prompt:

keepfire use --id <id> "current task text"     # any unique id fragment works
keepfire use --pick 2 "current task text"      # or pick the Nth candidate

The output is the working instruction — follow it to execute the coding task. No CLI? Read the card and adapt it yourself: current task + filled skeleton + constraints + output contract + "preserve these qualities" + "avoid".

  1. Close the loop. After the task succeeds, if the user says the result was better than before, offer to save an improved version:
keepfire improve <id> --note "what changed" --why "signal1|signal2"

Judgment calls

  • The recipe's structure is the value; its old context is disposable.
  • If two recipes half-match, prefer the one whose intent matches, then whose stack matches.
  • A recipe used successfully bumps its ranking automatically (use_count) — no action needed.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,645. 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.