skill-curator

Engineering Community

Repo stars 487
Author repo vibecosystem

Interpretation is structured for decision-making; original keeps the upstream SKILL.md unchanged.

Decide Fit First

Core job: Autonomous skill lifecycle manager. Reviews skill-compounder drafts, promotes high-confidence patterns, merges duplicates, archi…
Best fit: Use it when the task has reusable inputs, steps, and validation criteria rather than a one-off answer.
Avoid forcing it: If the source lacks commands, platform support, or external-service evidence, keep those fields unknown instead of guessing.

Design Intent

Structure: The skill is organized around “The Pipeline”, “Promotion Criteria”, “Archive Criteria”, “Duplicate Detection”, showing how the author expects the agent to judge fit, collect context, and produce verifiable output.
Trigger evidence: Prioritize the author’s wording around when to use it, what context to collect, and what output shape to produce.
Evidence boundary: Author text states facts, repository files prove commands and paths, and Fluxly only adds fit, limits, and usage judgment.

How To Use It

Inputs: Provide target material, scope, expected result, forbidden changes, and validation method.
Invocation: Name skill-curator directly; if the source includes slash commands, start with the command and then add task context.
Validation: Start small and check whether the result follows “The Pipeline / Promotion Criteria / Archive Criteria” before expanding.

Boundaries And Review

Dependencies: It usually needs no extra API key, so start with a small validation task.
Permissions: Declared permissions include read / write; ask the agent to state file, command, and rollback boundaries before acting.
Quality bar: A useful result names the deliverable, evidence, and next action. Generic prose means the task needs tighter context.

Skill Curator

Autonomous skill lifecycle system. Works in partnership with skill-compounder to close the loop: observe -> draft -> review -> promote.

The Pipeline

1. You work normally (no manual action needed)
2. skill-compounder (Stop hook) detects successful patterns
3. Draft written to ~/.claude/skills-drafts/
4. skill-curator (SessionStart hook) reviews all drafts
5. Drafts are either:
   - PROMOTED to ~/.claude/skills/ (confidence >= 80, no duplicates)
   - MERGED with existing similar skill
   - ARCHIVED to ~/.claude/skills-drafts/archive/ (confidence < 50)
   - KEPT in drafts for next review (50-79, needs more evidence)

Promotion Criteria

A draft is auto-promoted when ALL of these are true:

Criterion	Threshold
Confidence	>= 80
Occurrences	>= 3 (or category is error-recovery)
Uniqueness	Not duplicate of existing skill
Age	Less than 14 days old
Category	One of: repeated-workflow, error-recovery, domain-pattern

Archive Criteria

A draft is auto-archived when ANY of these are true:

Confidence < 50 (not reproducible)
Age > 14 days without promotion (stale)
Duplicate of existing promoted skill

Duplicate Detection

Before promoting, check against existing skills:

Name match - exact directory name
Category + description overlap - same category, 4+ keyword overlap

If duplicate found, either merge evidence into existing skill or archive.

Draft File Format

Drafts live in ~/.claude/skills-drafts/<name>.md:

---
name: pattern-abc123
description: "Procedural pattern: Read -> Edit -> Test -> Commit"
category: repeated-workflow
confidence: 85
occurrences: 7
source_session: s-xyz789
generated_at: 2026-04-12T03:45:00Z
status: draft
---

When promoted, status: draft becomes status: active.

Integration

skill-compounder (Stop hook): Generates drafts from session observations
self-learner: Provides error-pattern data that feeds skill-compounder
canavar: Cross-trains all agents when skill is promoted
layered-recall: Makes new skills queryable via Depth 1-3 pattern

Metrics

Tracked in ~/.claude/skill-evolution.jsonl:

draft_generated_at
promoted_at
time_to_promotion
usage_after_promotion
retention_rate (still active after 30 days)

Safety

No manual override needed - runs silently on session start
No destructive actions - drafts/skills are moved, not deleted
Rollback via git - all promoted skills are in git history
High threshold - 80+ confidence prevents low-quality promotion

Fluxly profile Author and license come from source; runtime, permissions, and network are Fluxly detections or estimates

Fluxly category: Engineering
Author-declared agents: No explicit declaration found; this is not inferred or tested compatibility
Static check: 88 / 100 · heuristic scan, not runtime safety proof
Author / version / license: @vibeeval · no license declared
Fluxly token estimate: Lean
Fluxly setup estimate: Plug-and-play
External API key: No requirement detected
Detected OS requirements: Unspecified
Runtime requirements: Unspecified
Detected file/system behavior: Read-only
Write / modify
Detected network behavior: Local-only
Install commands: None (reference only)

Profile is derived at build time from SKILL.md and install vectors. Subject to drift from author intent.

Heads up: 未限定 allowed-tools，默认拥有全部工具权限。

Output preview skill-curator.preview

The current SKILL.md does not define a fixed output example.

SKILL.md SKILL.md

# Skill Curator

Autonomous skill lifecycle system. Works in partnership with `skill-compounder` to close the loop: observe -> draft -> review -> promote.

## The Pipeline

```
1. You work normally (no manual action needed)
2. skill-compounder (Stop hook) detects successful patterns
3. Draft written to ~/.claude/skills-drafts/
4. skill-curator (SessionStart hook) reviews all drafts
5. Drafts are either:
   - PROMOTED to ~/.claude/skills/ (confidence >= 80, no duplicates)
   - MERGED with existing similar skill
   - ARCHIVED to ~/.claude/skills-drafts/archive/ (confidence < 50)
   - KEPT in drafts for next review (50-79, needs more evidence)
```

## Promotion Criteria

A draft is auto-promoted when ALL of these are true:

| Criterion | Threshold |
|-----------|-----------|
| Confidence | >= 80 |
| Occurrences | >= 3 (or category is error-recovery) |
| Uniqueness | Not duplicate of existing skill |
| Age | Less than 14 days old |
| Category | One of: repeated-workflow, error-recovery, domain-pattern |

## Archive Criteria

A draft is auto-archived when ANY of these are true:

- Confidence < 50 (not reproducible)
- Age > 14 days without promotion (stale)
- Duplicate of existing promoted skill

## Duplicate Detection

Before promoting, check against existing skills:
1. **Name match** - exact directory name
2. **Category + description overlap** - same category, 4+ keyword overlap

If duplicate found, either merge evidence into existing skill or archive.

## Draft File Format

Drafts live in `~/.claude/skills-drafts/<name>.md`:

```yaml
---
name: pattern-abc123
description: "Procedural pattern: Read -> Edit -> Test -> Commit"
category: repeated-workflow
confidence: 85
occurrences: 7
source_session: s-xyz789
generated_at: 2026-04-12T03:45:00Z
status: draft
---
```

…

Evidence boundary and execution chain

Author text anchors workflow facts; Fluxly only indexes current sections, terms, files, and commands.

sections -> The Pipeline → Promotion Criteria → Archive Criteria → Duplicate Detection → Draft File Format → Integration
terms    -> Name match · Category + description overlap · skill-compounder · self-learner · canavar · layered-recall · No manual override needed · No destructive actions
files/cmd -> skill-compounder · ~/.claude/skills-drafts/<name>.md · status: draft · status: active · ~/.claude/skill-evolution.jsonl
body sha256 -> 7d53a7d753af