Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/tzavislan/kumihimonpx agentmods add skills/tzavislan/kumihimo/kumihimo-iterationWrote 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/tzavislan/kumihimo/kumihimo-iteration)<a href="https://agentmods.dev/skills/tzavislan/kumihimo/kumihimo-iteration"><img src="https://agentmods.dev/badge/skills/tzavislan/kumihimo/kumihimo-iteration/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tzavislan/kumihimo/kumihimo-iteration"><img src="https://agentmods.dev/badge/skills/tzavislan/kumihimo/kumihimo-iteration.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00098 | $0.04046 |
| Opus 5 | $0.00049 | $0.02023 |
| Sonnet 5 | $0.00020 | $0.00809 |
| Haiku 4.5 | $0.00010 | $0.00405 |
Grade A, and why
kumihimo-iteration 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 5d 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.
How it starts
The opening of the file, as written. The whole thing — 312 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Kumihimo iteration — one pass
One invocation is exactly one iteration: one queue item, one verdict, one commit, one journal entry. A loop re-invokes this skill for the next item. Every iteration must leave the repository committed and clean enough for the next one to start cold.
Working directory is C:\Kumihimo. Nothing outside it is written except
Claude's own memory directory.
Before anything: the safety check
- Never
git push. Commits are local. Pushing publishes to the public repo and triggers CI; it happens at milestone close with Thomas's say-so, or when he asks. If a milestone completes mid-loop, propose the push in the report and keep looping. - Never publish to PyPI, create GitHub releases, or post anywhere.
- Never weaken an invariant in CLAUDE.md to make an item pass. If an item
and an invariant collide, the item is
needs-thomas.
Step 1 — Read the state
build/state/queue.md— the work, in order.build/state/journal.md— what the last few iterations did.build/state/loop.json— iteration counter and collision guard.
Collision guard. If loop.json has an active_session under 30 minutes
old that is not this session, another loop is running — stop and say so.
Otherwise claim it (session id + timestamp) and clear it when the iteration
ends.
Step 2 — Pick exactly one item
Take the first item with status todo whose dependencies are all done.
Do not batch. Do not skip ahead because something looks easier.
If an item is larger than one review round can hold, mark it split, write
the smaller items beneath it, and take the first.
Stop conditions — end the loop, do not spin:
- No
todoitems remain → say the queue is drained and stop. - Three consecutive iterations ended
blocked→ stop and report the pattern. - The same item has failed verification twice across two iterations → mark it
needs-thomas, write why, and stop. - The clean tree fails
uv run pytest -qbefore you changed anything → stop; the tree is broken and another iteration will not help.
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.
- 5d ago Changed · +86 lines 7504c89da370
- 9d ago First seen · 226 lines · 98 tokens per session scan A 3bdfe02a9ef5
kumihimo-iteration is a skill published in the GitHub repository tzavislan/kumihimo (0 stars, last pushed 6d ago), licensed MIT. It adds 98 tokens to every session and 4,046 once invoked, about $0.0005 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 2026-08-31.
Other skills, from other repositories
lean-commit
Ultra-compressed commit message generator. Cuts noise from commit messages while preserving intent and reasoning. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/commit", or invokes…
git-helper
Generate git commit messages and help with git workflows.
safe-data-export
Use when a task's output is sensitive or large data and the destination is a path inside a version-controlled tree — writing the data is not the finish line, an unreviewed commit of it is the failure mode. Triggers on "export the data", "dump the DB", "extract observations", "write it to the repo", "save the logs"…
tldr-commit
Ultra-compressed commit message generator. Cuts noise from commit messages while preserving intent and reasoning. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/commit", or invokes /tldr-commit.…
commit
Smart commit with Ollama-generated messages. Conventional commits, auto-push option. Triggers on: commit this, commit changes, smart commit, /commit.
contextual-commit
Write contextual commits that capture intent, decisions, and constraints alongside code changes. Use when committing code, finishing a task, or when the user asks to commit. Extends Conventional Commits with structured action lines in the commit body that preserve WHY code was written, not just WHAT changed.