agentleFS
exm7777exm7777/co-founder/.agents/skills/recall/SKILL.md

recall

Skill0 starsChanged 3 months agoMITView on GitHub

Check what the business already knows before deciding anything. Run when the user says "what do we know about...", "have we tried this before", "check the vault", "what did we learn about X", and AUTOMATICALLY before any gauntlet, plan, offer, or big decision, per the charter's recall-first ritual. Answering a business question from model memory when the vault holds the founder's real history is the failure this skill exists to prevent.

153 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use addskill, apply-firmware and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ryanportfolioSkill

    addskill

    Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.

    Also in 6 projects that use this

  • ryanportfolioSkill

    apply-firmware

    Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.

    Also in 6 projects that use this

  • +10 more from ryanportfolio

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 140 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/CONVENTIONS-core.md and run its Universal preflight.

    line 31
  2. Very specific:

    Read index.md first: it is the catalog, and it says where to drill.

    line 39
  3. Very specific:

    Before saying anything shaped like "we have no note on that": list the relevant wiki/ folders and search the question's key words (and …

    line 48
  4. Very specific:

    Read canonical lifecycle fields before using status words.

    line 71
  5. Distinctive:

    Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or …

    line 77
  6. Very specific:

    When recall writes a queue item, run scripts/graph-audit queue.md.

    line 103

“Common” is measured: kept by at least 6 in 10 of the 140 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

recall: the vault answers first

The most expensive sentence an AI can say to a founder is "we have no notes on that" when the notes exist. Recall reads before it speaks, cites what it finds by exact file, and admits plainly when the vault is genuinely empty.

  the question
       |
  [ index ]     read index.md first, drill into pages
       |
  [ verify ]    list + grep BEFORE any "nothing on record"
       |
  [ answer ]    cited by file, confidence surfaced
       |
  [ bank ]      a real synthesis goes back into the vault

Step 0: Preflight

Read references/CONVENTIONS-core.md and run its Universal preflight. A Fresh, Partial, or Wrong-folder result stops and runs the co-founder-setup skill before anything else (re-sync for Partial); an absent vault cannot be searched honestly. On Ready, read the charter's path map.

Done when: the founder system is Ready and the catalog path is known.

Step 1: Retrieve

Read index.md first: it is the catalog, and it says where to drill. Then read the pages that plausibly bear on the question, plus the matching fixed hub/<section>.md. Search wide before answering narrow: the useful lesson is often filed under the channel, the person, or the initiative rather than the topic word.

Done when: every plausibly relevant page has been looked at, not guessed at.

Step 2: The absence guard

Before saying anything shaped like "we have no note on that": list the relevant wiki/ folders and search the question's key words (and their obvious synonyms) across the vault with whatever the environment has (rg or grep where present, the harness's search or explicit listing plus reading otherwise, per CONVENTIONS). Absence gets VERIFIED, never remembered: answering from memory instead of searching is the single most common way systems like this fail. Genuinely empty: say exactly that, in one line, and offer the research skill to fill it.

Any durable absence statement or queue item records the shared verified-absence fields: paths or corpus, terms and synonyms, date, tool, and result count. Phrase only the scoped result. It can prove what is not recorded in the vault, never what the business or world has never done.

Done when: presence and absence are both grounded in an actual look.

Step 3: Answer with receipts

  • Cite the exact file for every claim: wiki/learnings/warm-intro-beats-cold.md, not "our notes." Run the shared durable claim admission gate: reopen the file, verify its current lifecycle and confidence, and quote the exact supporting passage. Trace derived claims to their original inputs. A page that calls a draft approved, or mentions a topic without the claimed evidence, does not pass merely because recall found it.
  • Surface confidence: a stated claim is repeated as a quote with its source, never as a truth. A low note gets its doubt said out loud.
  • Read canonical lifecycle fields before using status words. An offer is approved only at offer_state: approved or active; publication exists only at content_state: published with published_at. Contradictory index or hub prose is reported as drift, not repeated as truth.
  • Cross-file analogies and patterns pass the shared parameter gate. When no admitted receipt states the synthesis, label it Unverified synthesis: and offer the one founder question that could confirm it; rhetoric never upgrades itself into remembered fact.
  • Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or citing a receipt for it.
  • Notes past their review_after date get flagged as possibly stale.
  • Feeding another skill (gauntlet, plan, offer): deliver a task-scoped mini-brief, the 5-10 lines that bear on THIS decision, never a vault dump.
  • Exclude every type: content page from business-fact receipts. Drafts are proposals, and approved/published pieces are publication history, not the source of truth. Use wiki/business/offer.md for the current offer and wiki/business/metrics.md for current numbers.
  • This is the internal leg of the receipts law; when the question also needs external truth, say so and name research as the other leg.

Done when: the founder can click through to every fact in the answer.

Step 4: Bank real syntheses

A recall that produced genuine new synthesis (a comparison, a pattern across notes, a connection nobody had written down) is itself a finding. Append one pending entry to queue.md addressed from recall to steward. The request carries the exact synthesis and the source_paths list names every note it came from; recall reads, it never creates the wiki page. Narrate the queue ID in one line. A synthesis that evaporates into chat history has to be paid for again next month. Routine lookups do not get banked; the test is whether the answer contains something no single existing page says.

Mandatory filing gate

  1. When recall writes a queue item, run scripts/graph-audit queue.md.
  2. Fix every violation and rerun it. The queue item is not durable and no queue ID is narrated until it exits zero. A read-only lookup writes nothing and has no filing gate.

Done when: anything newly true has a durable pending queue ID, and everything else is cited.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.