agentleFS
mhxiemhxie/Atelier/skills/read/SKILL.md

read

Skill5 starsChanged 35 days agoMITView on GitHub

Read and discuss a source with one reading worker; expand perspectives, evidence gathering, or review when the task needs it.

19 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use hunt, think and 4 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.
  • ninehillsSkill

    hunt

    Finds root cause before applying fixes for errors, crashes, regressions, failing tests, broken behavior, and screenshot-reported defects. Use when users report in any language errors, crashes, broken behavior, regressions, failing tests, screenshot evidence, or something that used to work and now fails. Not for code review or new features.

    Also in 5 projects that use this

  • vltingSkill

    think

    Lightweight plan-to-execution pipeline. Explores codebase, runs council deliberation, presents plan for approval, then executes via /run.

    Also in 6 projects that use this

  • DockerSkill

    write

    Write or edit reader-facing technical prose for immediate comprehension. Use for documentation, PR titles and descriptions, release notes, design documents, user-facing explanations, and substantive comments. Also use when asked to make writing clearer, more natural, less AI-generated, or easier to read. For documentation fixes, handles edits, formatting, self-review, and commits after research identifies what to change. Do not use for code-only tasks with no prose deliverable.

    Also in 6 projects that use this

  • genomewalkerSkill

    health

    Soul system health check with remediation

    Also in 5 projects that use this

  • kt057Skill

    check

    lint, typecheck, test を一括実行して結果をサマリーする

    Also in 7 projects that use this

  • genomewalkerSkill

    learn

    Learn from web sources, generate SKILL.md, and store key patterns in soul memory. Trigger keywords: learn, skill-gen, web-learn, documentation, tutorial.

    Also in 7 projects that use this

The file

We split this skill into the rules it gives an agent

9 rules. 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Start with one Reader, or one Scholar when word_count > 8000, the source is under <paths.papers>/ or <paths.preprints>/, or frontmatter …

    line 19
  2. Very specific:

    Test for an already-cached PDF.

    line 35
  3. Very specific:

    After the reflection file is saved, make one readwise reader-create-document call only if the user explicitly authorized backing up this …

    line 52
  4. Very specific:

    If no unambiguous curation episode exists, use a new direct-reading episode with policy_id: direct.

    line 91
  5. Distinctive:

    Use the supplied source; ask only when the source or a requested focus is missing.

    line 105
  6. Distinctive:

    Add a specialist only for a concrete task the reading has exposed: Researcher for a needed connection to local notes, one Scout for bounded …

    line 116
  7. Distinctive:

    Use Synthesizer only when combining substantial independent briefs needs separate synthesis.

    line 120
  8. Distinctive:

    Follow the user's questions.

    line 132
  9. Distinctive:

    Every mode checks quotations and attributed claims against the source and distinguishes the author's claims from the reader's analysis.

    line 137

“Common” is measured: kept by at least 6 in 10 of the 14 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

Read & Discuss

Procedure for Read intent. Owns Reader/Scholar selection, Readwise prefetch, source backup, and the three read modes (Read & Discuss, Focused Read, Multi-Lens Read).

Menu

OptionLabelDescription
1Read & DiscussQuick read + interactive discussion (default)
2Focused ReadPick 1-2 specific lenses to focus on
3Multi-Lens ReadExamine the text through all four lenses

Reader vs Scholar selection (applies to all three Read modes)

Start with one Reader, or one Scholar when word_count > 8000, the source is under <paths.papers>/ or <paths.preprints>/, or frontmatter declares difficulty: hard. Both use the same reading workflow; their voice bindings differ. This procedure owns the selection rule.

Local cache check (before fetching; applies to all three Read modes when the source is a paper or external URL)

If the source is a paper or an external URL (arXiv, conference PDF, a paper named by title), check local material FIRST and only hit the web on a miss. The cached copy is often already on disk; a web round-trip before checking it is wasted latency.

  1. Surface prior local material. Run uv run scripts/semantic.py query "<title or distinctive keywords>" --top 5 --format json. The central corpus policy returns current authored notes and may return compact locators for assets under a raw/ cluster. It does not extract a binary PDF or index <paths.cache>/, so treat these hits as related context or a raw location hint, not as proof that the paper PDF is cached.
  2. Test for an already-cached PDF. Glob the flat paper store for a file matching the author or title: ls "$OV"/papers/ "$OV"/preprints/ 2>/dev/null | grep -i "<firstauthor-or-distinctive-keyword>" (resolve papers / preprints via harness/paths.toml). If a match exists, pass that local path to the reading agent and skip the web fetch entirely.
  3. Fetch from the web only if both the note query (step 1) and the PDF glob (step 2) miss. After a web fetch, cache the PDF into <paths.papers>/ so the next read is a local hit (naming convention in sources/local-papers.md).
  4. If the resolved source is a local PDF, materialize or reuse the paper text cache before dispatch. Run python3 scripts/paper_cache.py "<resolved-pdf-path>"; it returns <paths.cache>/<paper-slug>/. Pass that directory as cache_path to every Reader or Scholar. For a normal URL or a non-PDF source, skip this cache step and use the source directly. Follow the shared scratch rule in AGENTS.md; one-session page renders use mktemp -d and are removed when reading finishes.

Prefetch Step (Readwise podcasts, videos, articles; applies to all three Read modes)

If the source is a Readwise podcast, video, or article (user provides a Readwise URL, document_id, or names a podcast), cache the transcript once before dispatching any reading agent (Reader or Scholar). Independent fetches across parallel reading-agent instances are the failure mode this step exists to avoid (same reasoning as the paper cache).

  1. Resolve document_id. If the user gave a title, find it: readwise reader-search-documents --query "<keywords>" → pick the match.
  2. Snapshot content: readwise reader-get-document-details --document-id <id> | jq -r '.content' > <paths.cache>/rw-<id>.md
  3. Pass cache_path: <paths.cache>/rw-<id>.md to every reading-agent dispatch (Reader or Scholar). The convention is documented in agents/reader.md § "Readwise transcript cache"; Scholar follows the same convention.
  4. For podcasts specifically: also pass the guest name (parsed from title) and host name (from the author field) in the dispatch prompt so the reading agent doesn't have to re-infer for citation.

Backup to Readwise (optional final step in every Read mode)

After the reflection file is saved, make one readwise reader-create-document call only if the user explicitly authorized backing up this source to Readwise. Approval to save the local reflection does not authorize this external write. Without backup authority, skip it without blocking the local save. The reflection in <paths.reflections>/ remains the durable artifact.

Skip conditions (do NOT call the CLI):

  • The user has not explicitly authorized this Readwise backup.
  • Input was a Readwise URL or document_id (already in Readwise; the Prefetch Step handled it).
  • Input was a local [[Note Title]] (no source URL exists).
  • Input was a transcript paste with no accompanying URL (nothing to back up).
  • readwise CLI is not installed in the environment (command -v readwise returns nothing). Backup is best-effort; absence of the CLI is not a session error.

When it runs (orchestrator, not a separate agent):

readwise reader-create-document \
  --url "<canonical-source-url>" \
  --tags "<comma-list>" \
  --category "<article|pdf|video|podcast>"
  • --url: canonical source URL. For arXiv use the abs page (/abs/<id>), not the PDF URL.
  • --tags: 3-5 topic tags selected by the orchestrator within the authorized backup.
  • --category: default article; use pdf for arXiv/PDF papers, video or podcast for transcripts.

Print the resulting Readwise URL or document_id to the user as a one-liner confirmation. Do not pre-check for duplicates: if Readwise re-creates a doc, the second document_id is fine. Do not loop on errors; if the call fails (network, auth), report the error and continue, since the reflection file is already saved.

Initial-analysis persistence checkpoint

For a source handed off from Curate, retain its reading episode/item/policy IDs. Follow protocols/decision-ledger.md → Reading feedback loop for any explicit reading decision, consumption observation, or usefulness feedback that arrives during discussion. Record the user's reason and source turn; do not ask for a rating merely to complete telemetry. An analysis completing or a reflection being saved does not prove the user consumed or valued it. For direct reading, look up decisions.py reading-episodes --item <source-id> when attribution is needed. Use a prior episode only when its origin is unambiguous; a truncated or multiple-match result does not establish that. If no unambiguous curation episode exists, use a new direct-reading episode with policy_id: direct; its feedback may inform taste but receives no curation-policy outcome credit.

Immediately after the first complete reading analysis returns, run the reading checkpoint defined in protocols/session-log.md before presenting the analysis or entering discussion. Do not wait for a future write-back request. Resolve a collision-free reading session-log path, then create the complete log and its filled Reading Capsule in one Write/Edit operation. Do not use scripts/session_log.py for reading because skeleton-then-fill leaves an interruption window.

Per-option flows

Use the supplied source; ask only when the source or a requested focus is missing. Apply the matching cache/prefetch branch before dispatch.

  • Read & Discuss: One Reader or Scholar examines the argument and its evidence, starting with the Critical lens. Complete the Initial-analysis persistence checkpoint, present the analysis, and discuss it with the user.
  • Focused Read: The same worker applies the requested lens or lenses, keeping their findings distinct. Complete the same checkpoint before presentation or discussion.
  • Multi-Lens Read: On request, follow the Reading Hub flow below.

Add a specialist only for a concrete task the reading has exposed: Researcher for a needed connection to local notes, one Scout for bounded verification, Thinker for a consequential framework question, or Challenger for a specific assumption that needs independent scrutiny. Name that task in the dispatch. Use Synthesizer only when combining substantial independent briefs needs separate synthesis. Additional agents are not a completion requirement.

Reading Hub Flow (Multi-Lens Read)

  1. One Reader or Scholar applies Critical, Structural, Practical, and Dialectical perspectives in separate sections. Use independent readers when the user requests independent takes or a specific disputed claim warrants them; parallelize only those independent tasks.
  2. The orchestrator presents convergence, disagreement, and remaining uncertainty across the findings. Complete the Initial-analysis persistence checkpoint before presenting the report in Chinese or entering discussion.
  3. Follow the user's questions. Expand a lens or seek specific evidence when needed; there is no fixed fanout or mandatory synthesis stage.

Source verification and reflection writeback

Every mode checks quotations and attributed claims against the source and distinguishes the author's claims from the reader's analysis. Verify external claims with appropriate sources or leave them explicitly unverified. If the source cannot be accessed, state the limitation instead of reconstructing it.

Reading review: add one independent review for substantive, uncertain, or consequential claims, a significant disputed interpretation, or when the user asks. Dispatch the targeted Reviewer in Session Review mode and resolve findings before writeback. An ordinary reading reflection does not require a Reviewer/Challenger pair.

Present the proposed reflection and obtain explicit user approval before writing <paths.reflections>/YYYY-MM-DD-reading-<slug>.md. Discussion alone does not require a reflection file. Never write reading conclusions to daily notes. Include full source text only when supplied or locally retained with redistribution rights; otherwise preserve a source locator and bounded excerpts. After an approved reflection is saved, apply Backup to Readwise's separate authority and skip conditions above.

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.