Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/tpierrain/kenjaku/improvenpx skills add tpierrain/kenjaku --skill improvegit clone --depth 1 https://github.com/tpierrain/kenjakuWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tpierrain/kenjaku/improve)<a href="https://agentmods.dev/skills/tpierrain/kenjaku/improve"><img src="https://agentmods.dev/badge/skills/tpierrain/kenjaku/improve.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00031 | $0.00460 |
| Opus 5 | $0.00015 | $0.00230 |
| Sonnet 5 | $0.00006 | $0.00092 |
| Haiku 4.5 | $0.00003 | $0.00046 |
Grade A, and why
improve scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/improve — Continuous improvement of the harness
Routing
- No argument → Session mode (A).
ajouter <description>(oradd) → Quick-add mode (B).
MODE A — Improvement session
A1 — Read the backlog and observe
- Read
vault/backlog/harnais.md(backlog of pending improvements). - Read the most recently modified files in
.claude/skills/to understand the state of the harness. - If the conversation has 10+ exchanges, detect frictions:
- Repeated workarounds (same action 2+ times)
- Questions the vault couldn't answer
- Failed skills or unsatisfactory results
- Too many rounds to find a piece of information
Add the observations to the backlog with the
[observation]tag.
A2 — Propose the priority improvements
Present the 3 most impactful improvements, prioritized by impact/effort ratio, recurrence, leverage. For each: title + tag, impact (1 sentence), effort (quick win < 30 min / project), proposal (2-3 lines). End with: "Which one do we tackle?"
A3 — Implement
- Make the changes (skills, CLAUDE.md, vault structure…).
- Test if possible (dry run, syntax check).
- Mark the item
[x]+ date invault/backlog/harnais.md. - Let the hook commit.
MODE B — Quick add
Add the idea to vault/backlog/harnais.md: categorize (quick win / project / idea), [explicite] tag + today's date, confirm in one line. Do not implement.
Passive observation (outside the skill)
Rule in CLAUDE.md (section 5). Applies automatically to every session of 10+ exchanges, not only when
/improveis invoked.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 48 lines · 31 tokens per session scan A ef77fafb86f1
improve is a skill published in the GitHub repository tpierrain/kenjaku (80 stars, last pushed 10d ago), licensed Apache-2.0. It adds 31 tokens to every session and 460 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
lint
Run the LLM Wiki health check: orphans, broken links, contradictions, stale pages, index sync, MOC coverage, frontmatter fields, Exploration Gate, v5 verification fields, Bias Check, cross-vault links, and Core Context freshness; also responds to /lint and 위키 점검.
refresh-context
Re-snapshot Core Context.md from the 9 mothership system files and key essays when the mothership changes or the snapshot is older than 30 days; also responds to /refresh-context and 코어 컨텍스트 갱신.
status
Show the LLM Wiki state at a glance: counts, recent activity, index drift, inbox backlog, collectionPurpose/mainVault/explored/verificationStatus coverage, and Core Context snapshot age; also responds to /status and 상태 점검.
capture-tabs
Capture Chrome tab groups, browser research bundles, or ChatGPT/Gemini/Grok/Claude/Perplexity sessions into LLM Wiki Inbox markdown; also responds to 탭 캡처, 리서치 캡처, Deep Research 캡처, and /capture-tabs.
ingest
Ingest a URL, file, raw text, or Inbox item into the LLM Wiki with the purpose gate, mothership link search, raw-source preservation, Wiki compilation, index/log update, and review; also responds to /ingest and 인제스트.
reindex
Manually re-run qmd indexing and embeddings for wiki, rawsources, and queries after Obsidian edits, bulk moves, or embedding-model changes; also responds to /reindex and 재색인.