curate
Skill5 starsChanged 35 days agoMITView on GitHub
Goal-aware content curation and inbox triage with attributable reading episodes.
10 other projects use a version of this.
Adopted by
Most recently taken up by
- ai-erp-collab/deterministic-context-routing-plugintheir own version, first seen Oct 15, 2026
- arm092/autoharness-codextheir own version, first seen Oct 15, 2026
- MasRama/pekatheir own version, first seen Oct 13, 2026
- OHOKAYYEAH/claude-skillstheir own version, first seen Oct 13, 2026
- xiaolongnk/SuperHarnesstheir own version, first seen Sep 26, 2026
- njt2810/claude-code-frameworktheir own version, first seen Sep 10, 2026
Used with
Projects with it also use learn.
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.
genomewalkerSkilllearn
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 3 projects that use this
Other versions10 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 10
- iamneilroberts/claude-skills★ 26same name
- Reef123/OpenVera★ 9same name
- philippe-desplats/second-brain★ 6same name
- xiaolongnk/SuperHarness★ 4same name
- arm092/autoharness-codex★ 1same name
- MasRama/peka★ 1same name
- njt2810/claude-code-framework★ 1same name
- OHOKAYYEAH/claude-skills★ 1same name
- ai-erp-collab/deterministic-context-routing-plugin★ 0same name
- telefrek/vallorcine★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- atelierSkill
- autoevo-reviewSkill
- civSkill
- curateSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- lintSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
- captureSkill
- readingSkill
- autoevo-reviewSkill
- civSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- lintSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 10 edited versions we read keep these.
3
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
- Distinctive: line 24
Use only the packed route-selected profile files as goal and identity context.
- Very specific: line 26
Read protocols/decision-ledger.md → Reading feedback loop.
- Very specific: line 30
Read bounded explicit preferences with decisions.py reading-evidence.
- Distinctive: line 112
Ask: "Want me to tag these in Readwise and archive the rest?"
- Very specific: line 127
If the user wants to read something now, transition to /hi → Read with the selected item.
- Very specific: line 138
Create sources/<name>.md with setup, CLI commands, and output format
- Distinctive: line 139
Update the triage agent prompt to also pull from the new source
“Common” is measured: kept by at least 6 in 10 of the 10 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.
Files
Curate
Also reachable via
/hi <natural language>(e.g.,/hi triage inbox,/hi curate readwise,/hi score my inbox). Seeharness/intents.toml[intents.curate]for the row's example phrases. Both paths execute this same procedure.
Goal-aware content curation. Pulls from content sources, scores against your active goals and directions, and routes items into tiers.
Flow
1. Load Context (orchestrator)
Reuse the current curate context artifact from $hi; for direct
invocation, run:
uv run scripts/context_bundle.py --intent curateUse only the packed route-selected profile files as goal and identity context.
Read protocols/decision-ledger.md → Reading feedback loop. Capture the
selection policy and routed context with decisions.py reading-policy,
using the actual model that will perform triage. The helper stores replay
inputs privately once; retain its compact policy ID.
Read bounded explicit preferences with decisions.py reading-evidence; use
their reasons to refine selection, retaining uncertainty and context. Include
that evidence projection in the policy snapshot. Keep one episode ID for this
run and preserve each item's source ID through the reading handoff.
2. Dispatch Triage Agent (ad-hoc)
Dispatch a general-purpose agent (not a named agent) with this prompt structure:
You are triaging a content inbox against the user's active goals and directions.
## Goals and Directions
[paste relevant sections from profile/directions.md — current era, near-term goals, learning directions]
## Attributable Preferences
[paste the bounded reading-evidence projection, including reasons, uncertainty,
and proposed_action; approvals of archive actions are not positive taste]
Apply this feedback to selection when its context matches the present task.
## Task
1. Run BOTH commands in parallel (two Bash calls in one response):
- `readwise reader-list-documents --location new --limit 30 --response-fields title,author,summary,category,word_count,reading_time,saved_at,tags,source_url --json`
- `readwise reader-list-documents --location later --limit 20 --response-fields title,author,summary,category,word_count,reading_time,saved_at,tags,source_url --json`
Note: `id` is NOT a valid `--response-fields` value (the API rejects it). The document `id` is returned implicitly on every result as the top-level `id` key.
2. For each item, assign a tier. **Thresholds are commitment-adjusted by `category`.** A 69-minute podcast and a 1,200-word article are not the same ask:
| Category | deep-read means | digest | archive |
|---|---|---|---|
| `article`, `rss`, `email` | directly relevant to an active goal; worth ~10–30 min close read | interesting context, summary captures it | low relevance |
| `podcast`, `video` | relevant to an active goal AND `reading_time` ≤ ~90 min (the user will actually listen) | worth skimming highlights/summary only | too long or not relevant |
| `book`, `pdf`, `epub` | rarely auto-promoted; only if user explicitly tagged `#deep-read` | save summary, chapter-scan worthy | low relevance |
| `tweet` | almost never deep-read | quote worth keeping | archive |
3. Podcast metadata: the Readwise `author` field is the show **host**, not the guest. Parse the title for the guest. Common patterns:
- `"20VC: <Guest> on <Topic>"`
- `"<Show> with <Guest>: <Topic>"`
- `"Episode N: <Guest>, <Topic>"`
Cite the guest in the relevance reason, not the host (the host is the same every episode). Flag any auto-transcript as potentially misrendering proper nouns; do not write a guest name to the triage file without high confidence.
4. Write the ranked list to `<paths.cache>/triage-YYYY-MM-DD.md` with format:
```markdown
# Triage: YYYY-MM-DD
## Deep Read
- `id:DOC_ID` [Title](https://read.readwise.io/read/DOC_ID) (<category>, <commitment: e.g. "1h 9m listen" for podcasts or "~12 min read" for articles>, by <author or guest>): *one-line reason tied to a specific goal*
## Digest
- `id:DOC_ID` [Title](https://read.readwise.io/read/DOC_ID) (<category>, by <author or guest>): *one-line summary*
## Archive
- `id:DOC_ID` [Title](https://read.readwise.io/read/DOC_ID): *why skipped*
## Stats
- Inbox: N items | Later: N items
- Deep read: N | Digest: N | Archive: N
- Podcasts: N (total listen: Xh Ym) | Articles: N | Other: N- Return ONLY: the stats line + the Deep Read section (titles and reasons). Do not return the full list. Also return the complete structured candidate list as a local JSON artifact for the orchestrator: source ID, title, category, bounded source summary, canonical URL, author, reading_time, word_count, tags, proposed action, and reason. Preserve available selection inputs, omitting unavailable fields rather than fabricating them. This is proposal evidence.
### 3. Present Results (orchestrator)
Batch candidates as `proposed` with the captured policy ID using the typed
reading-event helper. Preserve stable event IDs for retries. After presentation,
batch `shown` events for the
items actually displayed. Internal archive candidates are not shown merely
because the agent wrote them to a file.
Show the user:
- Stats (inbox size, tier breakdown)
- Deep Read candidates with goal-relevance reasons
- Ask: "Want me to tag these in Readwise and archive the rest?"
### 4. Execute (orchestrator)
On approval:
- Record `approved` for the authorized operation on each identified item;
record any explicit per-item rejection/defer and its actual reason. Batch
approval does not establish that the user consumed or liked the content.
- Tag deep-read items: `readwise reader-add-tags-to-document --document-id <id> --tag-names deep-read`
- Tag digest items: `readwise reader-add-tags-to-document --document-id <id> --tag-names digest`
- Archive skipped items: `readwise reader-move-documents --document-ids <id1>,<id2>,<id3> --location archive`
- Tag all processed items: `readwise reader-add-tags-to-document --document-id <id> --tag-names auto-triaged`
### 5. Bridge to Reading (optional)
If the user wants to read something now, transition to `/hi → Read` with the selected item.
Carry its episode/item/policy IDs into Read; subsequent feedback keeps the
policy identity from selection time.
## Integration Points
- **`profile/reader_persona.md`** — if it exists, the triage agent can reference it for taste calibration alongside goals
## Adding New Sources
To add a new content source (e.g., Zhihu, Twitter):
1. Create `sources/<name>.md` with setup, CLI commands, and output format
2. Update the triage agent prompt to also pull from the new source
3. The same tier logic applies — deep-read / digest / archiveDoes 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
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.

