agentleFS
ARCASSystemsARCASSystems/FounderOS/skills/ingest/SKILL.md

ingest

Skill7 starsChanged 19 days agoMITView on GitHub

File a source (URL, file path, or pasted text) into raw/ with provenance preserved. Trigger on "ingest this", "process this source", "file this article", "save this transcript", "add this thread to the OS" (or run /founder-os:ingest). Also fires when the user shares a source they want preserved alongside extracted takeaways. Proposes wiki updates the user approves before writing. Different from knowledge-capture - ingest preserves the source.

88 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use daily-arxiv, exp-eval 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.
  • LambenthanSkill

    daily-arxiv

    每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新

    Also in 5 projects that use this

  • LambenthanSkill

    exp-eval

    实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges

    Also in 5 projects that use this

  • phanghonghaoSkill

    poster

    HTML/CSS A4 海报生成。从沙龙/活动 Markdown 文件生成科技风 A4 海报 PDF/PNG。Use when user mentions: poster, 海报, 海报生成, make poster, generate poster, 活动海报, 沙龙海报.

    Also in 5 projects that use this

  • skyllwtSkill

    exp-pilot-eval

    Pilot result evaluation — read pilot results, apply success criteria, update idea page (pilot_result, failure_reason if failed), generate PILOT_VERDICT_REPORT. Called by /ideate Phase 5 after /exp-pilot-run.

    Also in 4 projects that use this

  • skyllwtSkill

    exp-pilot-run

    Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

    Also in 4 projects that use this

  • +6 more from Lambenthan
  • +1 more from skyllwt

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. 0 are common, 7 are distinctive, and 8 are very specific to this project.

  • 0

    Common

    Most of the 72 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 8

    Very specific

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

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

    Run on a fresh install (creates raw/ lazily on first use).

    line 19
  2. Very specific:

    Never modify a file in raw/ after it's written.

    line 20
  3. Very specific:

    If core/identity.md does not exist, stop with: Founder OS not set up here.

    line 28
  4. Very specific:

    If raw/ directory does not exist, create it and copy templates/raw/README.md into it (so the convention is documented locally).

    line 29
  5. Distinctive:

    Anything else: - treat as pasted text. The argument body IS the source.

    line 37
  6. Very specific:

    If a file with that name already exists, append -2, -3, etc.

    line 61
  7. Distinctive:

    Write the file with this structure:

    line 63
  8. Distinctive:

    Read the source. Extract candidate updates across these categories:

    line 85
  9. Very specific:

    If there are zero candidates: No clear wiki updates surfaced.

    line 118
  10. Very specific:

    Use [[wiki-link]] syntax for all cross-references per the Wiki Conventions in CLAUDE.md.

    line 130
  11. Distinctive:

    Keep the proposal block scannable.

    line 157
  12. Distinctive:

    Never write to a wiki file without explicit user approval.

    line 158
  13. Distinctive:

    If the user declines all proposals, that's a valid outcome.

    line 159
  14. Very specific:

    Duplicate source: If a raw/ file with the same source_url already exists, ask: This source was ingested on <date>.

    line 163
  15. Distinctive:

    WebFetch fails: Ask user to paste the content.

    line 165

“Common” is measured: kept by at least 6 in 10 of the 72 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

Ingest

Runs on: local-writes - creates or edits files in your OS folder; needs an agent with write access.

Files a source into the OS with provenance. The raw layer is the immutable archive. The wiki layer holds extracted content with cross-links back to the source.

This skill must:

  • Auto-write only the raw source. Every wiki update requires explicit user approval.
  • Run on a fresh install (creates raw/ lazily on first use).
  • Never modify a file in raw/ after it's written.
  • Append a single line to brain/log.md capturing what was ingested and what wiki updates landed.
  • Render the proposal in under 3 seconds for sources under 50KB.

Pre-flight

  1. If core/identity.md does not exist, stop with: Founder OS not set up here. Run /founder-os:setup first.
  2. If raw/ directory does not exist, create it and copy templates/raw/README.md into it (so the convention is documented locally).

Step 1 - Resolve the source

Detect what the user passed in the argument:

  • URL (matches https?://) - fetch with WebFetch. If WebFetch fails, ask the user to paste the content directly.
  • File path (exists on disk per Glob) - Read the file content.
  • Anything else - treat as pasted text. The argument body IS the source.

Determine source_type from URL or content:

  • gist.github.com or raw markdown -> gist
  • youtube.com, podcast platforms, audio transcript -> podcast (or transcript if cleaner)
  • arxiv, pdf, paper -> article
  • twitter.com, x.com, linkedin.com posts, reddit threads -> thread
  • generic article URL -> article
  • file ending .pdf, .epub, .mobi -> book
  • otherwise -> document or other

Determine source_title:

  • For URLs: extract page title from fetched content. Fall back to URL slug.
  • For file paths: use the filename.
  • For pasted text: ask the user for a one-line title if not obvious from the first line.

Determine source_author:

  • For URLs: extract from byline/meta if found, otherwise omit.
  • Otherwise: omit unless obvious.

Step 2 - Write the raw file

Filename: raw/YYYY-MM-DD-<slug>.md where <slug> is a kebab-case 3-5 word version of the title.

If a file with that name already exists, append -2, -3, etc.

Write the file with this structure:

---
source_type: <type>
source_url: <url if applicable, omit if not>
source_title: <title>
source_author: <author if known, omit if not>
ingested_at: <YYYY-MM-DD>
ingested_by: ingest
wiki_pages: []
---

<the full source content>

wiki_pages is left empty for now. It gets updated in Step 6 with whatever the user approved.

Confirm to the user: Source filed to raw/<filename>. Now proposing wiki updates.

Step 3 - Extract candidate wiki updates

Read the source. Extract candidate updates across these categories:

CategoryTarget fileWhat qualifies
Personcontext/clients.md (append a roster row)A person worth tracking in the roster. Check context/clients.md first; append a new row only if not already there.
Decision inputcontext/decisions.md (append)Information that bears on an open or pending decision. Check current context/decisions.md for matches.
Patternbrain/patterns.md (append)A recurring pattern, framework, concept, or principle worth remembering.
Action itemcadence/daily-anchors.md or cadence/weekly-commitments.md (append to Could Do)A concrete next action surfaced by the source.
Companycompanies/<slug>-business.md (operator path, append a Source notes section) OR companies/prospects/<slug>.md (prospect path, append a Notes entry)Information about a company or organisation. Check the operator path first, then the prospect path. If no file exists at either path and the company is one the user runs, propose creating the operator file with a one-line description. If the company is one the user is selling to or watching, propose running the prospect-init flow instead.

Aim for 2-5 candidates total. Do NOT propose more than 7. If a source produces zero clear candidates, that's fine - proceed to Step 4 with an empty list.

Step 4 - Show proposal to user

Output the proposal as a single fenced block:

INGESTED: <title>
Source: raw/<filename>

Proposed wiki updates (you approve each):

[1] <category>: <target file>
    <one-sentence summary of the proposed addition>

[2] <category>: <target file>
    <one-sentence summary>

...

Reply with: yes (all), or list numbers to keep (e.g. "1,3"), or no (skip all).

If there are zero candidates: No clear wiki updates surfaced. Source filed to raw/<filename> for later reference. Then jump to Step 6 with an empty approval list.

Step 5 - Apply approved updates

For each approved update:

  • Person (append): Append a roster row to context/clients.md with <name> - <one-line note> plus Per [[raw/<filename>]].
  • Decision input (append): Append to context/decisions.md under the relevant decision header with provenance: Per [[raw/<filename>]]: <update>.
  • Pattern (append): Append to brain/patterns.md: ## YYYY-MM-DD - <pattern title> plus body and [[raw/<filename>]] reference.
  • Action item (append): Append to the relevant cadence file's "Could Do" section with [[raw/<filename>]] reference.
  • Company (append or create): Check companies/<slug>-business.md first (operator path). If it exists, append to a "Source notes" section with [[raw/<filename>]] reference. If the operator file does not exist, check companies/prospects/<slug>.md (prospect path) and append to its Notes section with the same reference. If neither file exists, propose creating the operator file (for companies the user runs) or running the prospect-init flow (for companies the user sells to or watches) before appending. Do not write both files in one pass - the source belongs at one path, not both.

Use [[wiki-link]] syntax for all cross-references per the Wiki Conventions in CLAUDE.md.

Step 6 - Update raw frontmatter and brain log

  1. Edit the raw file's frontmatter to fill wiki_pages with the actually-written list. Empty list if user declined all.
  2. Append a single line to brain/log.md:
    ## YYYY-MM-DD ingest | <source title> | wiki updates: <list of files written, or "none">

Step 7 - Output summary

Five-line confirmation:

INGESTED: <title>
RAW: raw/<filename>
WIKI UPDATES: <count> (<list of files>)
LOG: brain/log.md updated
NEXT: Run /founder-os:lint anytime to audit cross-references.

Output rules

  • Keep the proposal block scannable. One line per candidate plus one body line.
  • Never write to a wiki file without explicit user approval. raw/ is the only auto-write.
  • If the user declines all proposals, that's a valid outcome. The source still got filed.

Edge cases

  • Duplicate source. If a raw/ file with the same source_url already exists, ask: This source was ingested on <date>. Re-ingest with a fresh proposal? Default: no.
  • Very large source (>200KB). Truncate content in the raw file to the first 200KB and append ... [truncated, original at <url>]. Note in proposal that extraction was truncated.
  • WebFetch fails. Ask user to paste the content. Do not make up content.
  • Source has no extractable structure (e.g., a tweet, a one-line note). Filing to raw/ is still valuable. Output proposal: No clear wiki updates from this source. Filed for reference.
  • User says "yes" to a category but the target file is missing. Create the target file first, then write the update. Mention the file creation in the summary.
<!-- private-tag: not applicable: ingest writes raw source provenance and structured wiki updates; sources are URLs or files, not private speech -->

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.