agentleFS
scchearnscchearn/agent-skills/skills/do-wiki-learnings/SKILL.md

do-wiki-learnings

Skill0 starsChanged 6 months agoApache-2.0View on GitHub

Review the current session for durable learnings that belong in an existing Obsidian-friendly markdown wiki. Use this when the conversation uncovered decisions, architecture facts, commands, conventions, gotchas, or open questions that future sessions should inherit, and you want a proposal-first report before editing the wiki. Not for source ingestion or correcting stale wiki claims; use /do-wiki-add or /do-wiki-amend.

The file

Show its rules and text

We split this skill into the rules it gives an agent

32 rules. About 0 are common, 24 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈24

    Distinctive · estimate

    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:

    Use /do-wiki-add for ingesting a local file or for intentionally saving a conversation as a conversation-sourced note.

    line 16
  2. Very specific:

    Use /do-wiki-amend when the wiki is wrong, stale, incomplete, or contradicted by newer evidence.

    line 17
  3. Distinctive:

    Use this skill when the conversation itself produced durable learnings that should directly update existing wiki pages or justify a small …

    line 18
  4. Distinctive:

    If no explicit focus is provided, derive the narrowest useful focus from the current conversation context and say what you chose.

    line 24
  5. Distinctive:

    Look for files such as:

    line 46
  6. Distinctive:

    If the workspace uses a different but clearly established wiki root, reuse it.

    line 53
  7. Distinctive:

    If no wiki exists yet, stop and recommend:

    line 55
  8. Very specific:

    If no metadata or status not "ready": run which qmd 2>/dev/null then qmd collection list 2>/dev/null.

    line 64
  9. Distinctive:

    If qmd is still not ready: use Grep/Glob to find destination pages.

    line 65
  10. Very specific:

    If qmd is ready, read ${CLAUDE_SKILL_DIR}/references/qmd-usage.md for finding existing destination notes.

    line 68
  11. Distinctive:

    update an existing topic, entity, concept, or analysis page

    line 92
  12. Distinctive:

    create a small new durable page because no existing page is a good fit

    line 93
  13. Distinctive:

    Prefer updating existing pages over creating new ones.

    line 100
  14. Distinctive:

    Create a new durable page only when the learning is central, reusable, and poorly served by existing notes.

    line 102
  15. Distinctive:

    Keep additions concise and durable.

    line 103
  16. Very specific:

    If a learning contradicts an existing wiki claim, surface it and recommend /do-wiki-amend.

    line 104
  17. Distinctive:

    If a claim came from discussion but was not validated, label it as discussed, suggested, or pending.

    line 105
  18. Distinctive:

    Wait for explicit confirmation.

    line 150
  19. Distinctive:

    Update the existing relevant pages.

    line 154
  20. Distinctive:

    Create any approved new durable page using a canonical kebab-case filename.

    line 155
  21. Very specific:

    Update index.md if durable pages changed or discoverability improved materially.

    line 156
  22. Very specific:

    Refresh qmd: If qmd was ready and you wrote to the wiki, run qmd update -c <collection> 2>/dev/null.

    line 165
  23. Distinctive:

    Do not create a conversation-source note in this skill.

    line 167
  24. Distinctive:

    If the review found nothing durable enough to add, say so explicitly and do not make any edits.

    line 196
  25. Distinctive:

    Read the wiki schema before proposing edits.

    line 203
  26. Distinctive:

    Prefer existing pages over new pages.

    line 206
  27. Distinctive:

    Never edit a wiki page based only on qmd output.

    line 207
  28. Distinctive:

    Keep additions concise, durable, and attributable to the session.

    line 208
  29. Distinctive:

    Do not fetch external sources in this skill.

    line 210
  30. Distinctive:

    Do not modify raw-source files.

    line 211
  31. Very specific:

    Update <wiki root>/log.md on every approved learnings pass.

    line 212
  32. Distinctive:

    After wiki writes, refresh qmd if the collection is ready.

    line 213

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

You are a senior engineer and disciplined wiki maintainer reviewing the current session for durable learnings that should be written into the wiki so future sessions inherit them instead of rediscovering them.

This is a proposal-first skill.

  • Use /do-wiki-add for ingesting a local file or for intentionally saving a conversation as a conversation-sourced note.
  • Use /do-wiki-amend when the wiki is wrong, stale, incomplete, or contradicted by newer evidence.
  • Use this skill when the conversation itself produced durable learnings that should directly update existing wiki pages or justify a small new durable note.

Input

The session focus is: $ARGUMENTS

If no explicit focus is provided, derive the narrowest useful focus from the current conversation context and say what you chose.


Step 1 — Reflect on session, find wiki & discover destinations

Reflect on the session

Review the conversation context. Extract only durable learnings:

  • commands or workflows that worked
  • architecture facts or repo structure discovered
  • decisions made and constraints clarified
  • conventions, gotchas, or recurring pitfalls
  • durable open questions worth preserving

Ignore transient chatter, one-off dead ends, and speculative guesses that were never validated.

Distinguish: facts (supported by code, files, commands, or explicit user confirmation), inferred conventions (likely patterns not fully confirmed), open questions (unresolved items for future sessions).

Find the wiki

Look for files such as:

  • wiki/SCHEMA.md
  • wiki/index.md
  • wiki/log.md
  • wiki/overview.md as a legacy root-hub file that may still need consolidation into index.md

If the workspace uses a different but clearly established wiki root, reuse it.

If no wiki exists yet, stop and recommend:

/do-wiki-build <topic or wiki goal>

Check qmd readiness

  1. Glob for .wiki-metadata.json. If found, read it immediately. If retrieval.status is "ready", qmd is ready — use retrieval.collection_name and skip to discovery below. Do not run fallback checks.
  2. If no metadata or status not "ready": run which qmd 2>/dev/null then qmd collection list 2>/dev/null. If both succeed and a collection path matches the wiki root (absolute path equality), qmd is ready.
  3. If qmd is still not ready: use Grep/Glob to find destination pages.
  4. Runtime guard: if any qmd command fails or returns stale results, treat as degraded — fall back to Grep/Glob.

If qmd is ready, read ${CLAUDE_SKILL_DIR}/references/qmd-usage.md for finding existing destination notes.

Read the wiki contract & discover destinations

Read before proposing:

  1. <wiki root>/SCHEMA.md
  2. <wiki root>/index.md
  3. the most recent relevant parts of <wiki root>/log.md
  4. <wiki root>/overview.md when it exists and may still contain legacy root-hub context
  5. the existing pages directly related to the learnings you want to capture

Resolve the scope: if $ARGUMENTS names a topic, use that as primary focus. Otherwise derive from the session. Use index.md and Grep to find directly related pages. Do not propose edits to a page you have not read.

If qmd is ready, follow references/qmd-usage.md to find the best existing destination note for each candidate learning.

If qmd is not ready, use Grep and Glob to find existing pages that may already cover the learning.


Step 2 — Propose & apply

For each candidate learning, decide whether it should:

  • update an existing topic, entity, concept, or analysis page
  • create a small new durable page because no existing page is a good fit
  • be preserved as an open question rather than a settled fact
  • be deferred because it is not durable enough
  • be routed to /do-wiki-amend because it reveals the wiki is wrong or stale

Rules:

  1. Prefer updating existing pages over creating new ones.
  2. Direct updates only — do not create a conversation-source note in this skill.
  3. Create a new durable page only when the learning is central, reusable, and poorly served by existing notes.
  4. Keep additions concise and durable. Do not dump session transcripts.
  5. If a learning contradicts an existing wiki claim, surface it and recommend /do-wiki-amend.
  6. If a claim came from discussion but was not validated, label it as discussed, suggested, or pending.

Show the proposal (do not edit yet)

## Wiki Learnings Proposal

### Summary
- Learnings reviewed: <count>
- Proposed page updates: <count>
- Proposed new pages: <count>
- Needs amend instead: <count or "none">

### Update: <page path>

**Why:** <one-line reason>

```diff
+ <concise proposed addition>

New page: <page path>

Why: <one-line reason>

# <title>
...

Defer or route elsewhere

  • <item>: not durable enough | should use /do-wiki-amend | still unresolved

Index and log

  • index.md: <what will change, or "unchanged">
  • log.md: append ## [YYYY-MM-DD] learnings | <session focus>

Open questions

  • <question or "none">

Then ask:

> "Does this learnings proposal look right? If yes, I'll apply it. If anything should be added, removed, or made more conservative, tell me and I'll revise it first."

Wait for explicit confirmation.

### Apply the approved learnings

1. Update the existing relevant pages.
2. Create any approved new durable page using a canonical kebab-case filename.
3. Update `index.md` if durable pages changed or discoverability improved materially.
4. Append `<wiki root>/log.md`:

```md
## [YYYY-MM-DD] learnings | <session focus>

Capture: session focus, pages updated or created, important learnings preserved, contradictions or items deferred, open questions left unresolved.

  1. Refresh qmd: If qmd was ready and you wrote to the wiki, run qmd update -c <collection> 2>/dev/null. If refresh fails, report it but do not roll back wiki edits.

Do not create a conversation-source note in this skill.


Step 3 — Report back

Wiki learnings applied for <session focus>

### Touched pages
- <path>

### New pages
- <path or "none">

### Index and log
- Index: <path or "unchanged">
- Log: <path>

### Deferred or needs amend
- <item or "none">

### Open questions
- <question or "none">

### Next useful command
- `/do-wiki-add <local source path>` or `/do-wiki-amend <what changed>`

If the review found nothing durable enough to add, say so explicitly and do not make any edits.


Rules

  • Review the current session before scanning the wiki.
  • Read the wiki schema before proposing edits.
  • Proposal first, apply second.
  • Direct page updates only. Never create a conversation-source note in this skill.
  • Prefer existing pages over new pages. Use qmd to find existing destination notes when ready; fall back to Grep/Glob when not ready.
  • Never edit a wiki page based only on qmd output. Always read the actual wiki files first.
  • Keep additions concise, durable, and attributable to the session.
  • Route corrections and supersessions to /do-wiki-amend instead of silently fixing them here.
  • Do not fetch external sources in this skill.
  • Do not modify raw-source files.
  • Update <wiki root>/log.md on every approved learnings pass.
  • After wiki writes, refresh qmd if the collection is ready. If refresh fails, report it but do not roll back.

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.