Kihagyás

Skill-distill candidates — 2026-05-17#

SV B-4 + B-8 cross-axis sprint, skeleton mode (Week 1). Sessions scanned: 72 · with events: 24 · min-count threshold: 5

Top repetitive tool / skill tokens#

# Token Count Distinct sessions Examples
1 batch 15 6 2026-05-08-rojt-s-bojt-weboldal, 2026-05-11-github-repo, 2026-05-12-obsidian-vault, +3
2 ingest 14 4 2026-05-12-obsidian-vault, 2026-05-16-kgc-robbantott-brakeres, 2026-05-17-obsidian-vault, +1
3 backfill 9 5 2026-04-30-vault-restructure, 2026-05-13-sv-functional-payoff, 2026-05-17-obsidian-vault, +2
4 vault-cleanup 7 4 2026-04-23-myforge-dashboard, 2026-05-09-vault-maintenance, 2026-05-12-obsidian-vault, +1
5 vault-search 6 5 2026-05-13-sv-b2-memory-architecture, 2026-05-13-sv-week1-implementation, 2026-05-14-kgc-marketing, +2

Top tool-sequence bigrams (composition signal)#

# A → B Count
1 vault-cleanup → vault-cleanup 3
2 batch → batch 3
3 ingest → ingest 3
4 backfill → backfill 2
5 batch → backfill 2
6 backfill → ingest 2

Candidate distill targets#

Tokens crossing threshold are candidates for auto-skill distillation. Skeleton mode does NOT generate drafts — Week 2 will spawn an LLM agent to draft ~/.claude/skills/auto-distilled/queue/<slug>.md per candidate.

  • [1] batch — 15 uses across 6 sessions — 🟢 strong
  • [2] ingest — 14 uses across 4 sessions — 🟢 strong
  • [3] backfill — 9 uses across 5 sessions — 🟡 borderline
  • [4] vault-cleanup — 7 uses across 4 sessions — 🟡 borderline
  • [5] vault-search — 6 uses across 5 sessions — 🟡 borderline

Safety#

  • This run ONLY produced this audit file. No writes to ~/.claude/skills/.
  • Week 2 drafts will go to ~/.claude/skills/auto-distilled/queue/ (REVIEW state).
  • Activation: user manually runs mv queue/<slug>.md auto-distilled/<slug>/SKILL.md.
  • Forbidden targets: anywhere outside ~/.claude/skills/auto-distilled/.