tomevault-io Origin Evolve
Mission: turn .claude/traces/trace.md into approved updates of .claude/skills/.
Trigger: user input origin evolve (or equivalent intent).
Quick Navigation
| Need | See |
|---|---|
| Full proposal examples | EXAMPLES.md |
| Evidence requirements (Rule 1) | SKILL_MEMORY.md#rule-1-evidence-based-proposals |
| Attribution decision tree (Rule 2) | SKILL_MEMORY.md#rule-2-attribution-decision-tree |
| Append / Merge / Retire operations (Rule 3) | SKILL_MEMORY.md#rule-3-append--merge--retire |
| User approval policy (Rule 4) | SKILL_MEMORY.md#rule-4-user-approval-required-for-every-write |
| Format & capacity compliance (Rule 5) | SKILL_MEMORY.md#rule-5-format--capacity-compliance |
| Common pitfalls | SKILL_MEMORY.md#pitfalls |
| When to update this skill | ITERATION_GUIDE.md |
Trace Fields
Hook-generated (read-only): timestamp, modules, files_modified, file_count, lines_changed, edit_count, score, correction.
AI-supplemented: type, skills, mode, request, intent.
Lifecycle: status (pending/processed/expired/invalid), validated (_/true/false/pending-pipeline/invalid).
Execution Flow
Step 1 Read & Triage -> Step 2 Identify Patterns -> Step 3 Generate Proposals -> Step 4 User Approval -> Step 5 Write & Mark Processed -> Step 6 Calibrate (optional)
Step 1: Read & Triage
Acquire .trace_lock (overwrite if stale). Apply lifecycle transitions: pending with stale validated -> expired; pending-pipeline past expiry -> invalid.
Schema version gate: verify all pending entries have schema:1 in the TRACE header. Entries without a schema tag are treated as schema 1 (backward compatible). If any entry has schema:N where N > 1, abort and report "Unsupported trace schema version N. Update origin-evolve."
Read trace.md, keep pending only, then exclude entries with validated:pending-pipeline from proposal candidates until they are finalized to true / false or expire to invalid. If fewer than 5 analyzable pending entries remain and no correction signals exist, suggest waiting.
Compute three diagnostic counts across .claude/skills/*/SKILL_MEMORY.md and include in the analysis summary:
- within-skill rule pairs with anchor Jaccard >= 0.5
- cross-skill identical anchor sets
- cross-skill rule pairs with anchor Jaccard >= 0.5
Non-zero counts indicate prior attribution or merge errors and should inform Step 2 proposal generation.
Sort priority:
- Exclude
validated:pending-pipelinefrom the priority queue; they are runtime-in-flight, not proposal evidence. - P0:
validated:false(sub-sort: auto:major > auto:minor >_) - P1:
validated:true+ correction:auto:major - P2:
validated:true+ correction:auto:minor - P3:
validated:_with any correction signal - P4:
validated:_, no correction (by score desc)
Step 2: Identify Patterns
Focus on high-leverage signals; require 3+ supporting traces:
- Correction cluster — same module, multiple
correctionentries -> module lacks rule guidance - Module hotspot — same module pair co-occurring -> undocumented cross-module dependency
- Complexity concentration — single file with
edit_count>= 10 across multiple traces -> missing usage examples - Cross-skill overlap signal — Step 1 diagnostic counts non-zero -> attribution review candidate
Speculative patterns (use only with overwhelming evidence): knowledge gap (skills:[]), semantic drift (validated:false + correction:_), IDP gap (mode:_ dominant).
Step 3: Generate Proposals
For each pattern, produce a proposal containing:
- Operation: Append, Merge, or Retire (Rule 3)
- Target skill and file (Rule 2)
- Full content with Anchors and Related fields
- Evidence: timestamps + modules of supporting traces
- Risk note and confidence
Pre-write check: capacity headroom; for Retire, grep evidence that anchors are absent from current code.
When module-list and anchor evidence disagree on attribution, present BOTH candidate skills in Step 4 for user choice.
Step 4: User Approval
Present proposals one at a time:
- Append — full new entry
- Merge — original, merged version, and diff
- Retire — content + grep verification +
[RETIRED]effect
Rejection records an EVOLVE_REJECTION entry with pattern name, reason, and future scope effect.
Step 5: Write & Mark Processed
Atomic write (temp + rename). Replace analyzed entries with one audit line:
<!-- PROCESSED ts:{ISO8601} entries:{N} proposals:{M} -->
Delete .trace_lock in finally block.
Step 6: Calibrate (Optional)
When 20+ entries processed in this run, compare F/D/K/S/E dimension means between proposal-yielding traces and non-yielding traces. Adjust the dimension with the largest gap by 5-10% (weights 0.2-3.0, thresholds 1.0-3.0). At most one dimension per run.
Source: yunzeforbetter/CastFlow — distributed by TomeVault.