agentleFS
QuarktexQuarktex/qcobrain/.claude/skills/capture/SKILL.md

capture

Skill4 starsChanged 4 months agoMITView on GitHub

Ingest one raw input — a URL, video, article, transcript, screenshot text, meeting note, or a thought — into your co-brain. Saves the raw material verbatim to `sources/` (+ a `_sources.md` row + content hash), then writes atomic, fully-sourced notes into `wiki/`. Anything worth keeping but not verbatim-sourceable lands in `inbox/` with `status: needs-source`; junk is dropped per the blocklist. This is the WRITE side of cite-or-refuse. Triggers - "capture this", "add to my brain", "remember this", "ingest", or pasting something worth keeping.

60 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use weekly, ingest and 9 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.
  • gaiaaiagentSkill

    weekly

    Generate a weekly digest rolling up dailies into higher-level pattern analysis

    Also in 4 projects that use this

  • JaimeGBASkill

    ingest

    Digest an external source (article, paper, transcript, URL, book chapter) and integrate its content into the knowledge base

    Also in 8 projects that use this

  • AnthropicSkill

    brief

    Generate contextual briefings for legal work — daily summary, topic research, or incident response. Use when starting your day and need a scan of legal-relevant items across email, calendar, and contracts, when researching a specific legal question across internal sources, or when a developing situation (data breach, litigation threat, regulatory inquiry) needs rapid context.

    Also in 6 projects that use this

  • QuixiAISkill

    weekly-review

    Review the past week across goals, tasks, calendar, journal, and commitments, then propose the next week

    Also in 5 projects that use this

  • dubwoahSkill

    ask

    Answer questions about the codebase without generating or modifying any code. Use when the user wants to understand how something works, find where something is defined, or explore the project structure. Invoked with /ask.

    Also in 4 projects that use this

  • AlexPEClubSkill

    refine

    Always use when the user wants to discuss an exsiting feature or specification. Open an existing feature spec to improve, extend, or fundamentally challenge it. Pass the feature ID as argument (e.g. /refine PROJ-2).

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 0 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 58 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

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

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

    Before anything else, check the input against references/blocklist.md.

    line 18
  2. Very specific:

    Write the raw material to sources/<source-key>.md.

    line 25
  3. Very specific:

    Add (or update) the row in sources/_sources.md: source-key | path | sha256[:12] | origin | captured | last_verified.

    line 29
  4. Very specific:

    Write each new note to wiki/<id>.md per references/wiki-spec.md:

    line 36
  5. Very specific:

    Never invent or guess a source: No verbatim quote → inbox/, not wiki/.

    line 66
  6. Very specific:

    Provenance bottoms out in sources/: Never cite a wiki/ note as a source.

    line 68
  7. Very specific:

    Blocklist is law: When in doubt, quarantine to inbox/.

    line 72

“Common” is measured: kept by at least 6 in 10 of the 58 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

Inherits references/kernel.md (the provenance contract). Reads references/wiki-spec.md for schema.

What this does

Turns one raw input into (a) one sources/ capture file — the bottom of the provenance chain — and (b) one or more atomic, source-cited notes in wiki/. Anything that can't be sourced verbatim goes to inbox/. Anything on the Do-NOT-capture blocklist is dropped, never written.

Execution

Step 0 — Triage against the blocklist

Before anything else, check the input against references/blocklist.md. Transient/env-specific errors, negative tool-capability claims ("X doesn't work"), one-off chatter, secrets/PII not meant for canon, and speculation stated as fact get dropped — they poison the brain if persisted. When in doubt → inbox/, not wiki/.

Step 1 — Land the raw source

  1. Identify the input and its real-world origin (a URL, a video link, "meeting {{date}}", a file path, etc.). If the input is a URL, fetch it. If it's a thought with no external origin, origin = self / {{you}}, <date>.
  2. Write the raw material to sources/<source-key>.md. Source-key = short kebab slug, date-suffixed if it could collide (e.g. ai-os-talk-2026-06-23). Preserve text verbatim — this is the load-bearing copy. Keep a one-line header with origin + captured date. Type-folders are allowed (sources/youtube/, sources/meetings/).
  3. Compute the hash: shasum -a 256 sources/<source-key>.md | cut -c1-12.
  4. Add (or update) the row in sources/_sources.md: source-key | path | sha256[:12] | origin | captured | last_verified.

Step 2 — Extract atomic notes

  1. Re-read the source file back from disk — never write a quote from memory.
  2. Break the content into atomic claims — one idea each. Split compound ideas ("X and Y" → two notes).
  3. For EACH claim, before creating a note: grep INDEX.md for an existing note on the same idea. If one exists, refine it (add a src block) instead of forking a near-duplicate.
  4. Write each new note to wiki/<id>.md per references/wiki-spec.md:
    • frontmatter — id, title, type, tags, confidence, status, origin, pinned, created, source_verified_on, links;
    • a short interpretive body in your voice (see references/voice.md);
    • inline [[wikilinks]] to related notes, mirrored into the links: array (wikilinks are navigation only, never provenance);
    • one src block per claim, each quote a verbatim substring of the source file. Grep the source to confirm the quote resolves verbatim before saving.
  5. Set confidence: stated (the source asserts it directly) | inferred (you concluded it across spans; then attach all the spans as multiple src blocks) | reported (the source is relaying a third party).
  6. Set origin: user-directed when a human directed this capture (the default for /capture). An autonomous loop sets agent-generated. Loops may later auto-manage only agent-generated notes; user-directed and pinned: true are immune.

Step 3 — Quarantine what you can't source

  1. A claim worth keeping but lacking a verbatim source → write inbox/<slug>.md with status: needs-source, an origin: line, and a note on what source would confirm it. Never smuggle it into wiki/.

Step 4 — Third-party claims (the attribution layer)

  1. A fact that belongs to another party enters your canon ONLY as an attributed, sourced claim — frame the body as "per {{Acme}}, as of {{date}}", set confidence: reported, and cite the source where they stated it. Their knowledge feeds your brain without becoming indistinguishable from your own.

Step 5 — Update the index

  1. Append the new notes to INDEX.md (or run /wiki-doctor to rebuild it deterministically). Add any obvious natural-language query aliases to the ALIASES section.
  2. Close with a one-screen summary: source-key created, N notes added, M drafts quarantined, J dropped, and any contradictions spotted vs. existing notes.

Guardrails (hard)

  • Never invent or guess a source. No verbatim quote → inbox/, not wiki/.
  • Quotes are verbatim substrings of the resolved source file. Grep to confirm before saving.
  • Provenance bottoms out in sources/. Never cite a wiki/ note as a source.
  • Cap quotes (~300 chars). Interpret in the body; don't paste transcripts into wiki/.
  • Dedup first. Grep INDEX.md before creating; refine an existing note rather than fork it.
  • One idea per note. Compound titles get split.
  • Blocklist is law. When in doubt, quarantine to inbox/; never persist the forbidden classes.

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.