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 skills add gigsmart/haiku-method --skill haiku-repairgit clone --depth 1 https://github.com/gigsmart/haiku-methodWrote 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/gigsmart/haiku-method/haiku-repair)<a href="https://agentmods.dev/skills/gigsmart/haiku-method/haiku-repair"><img src="https://agentmods.dev/badge/skills/gigsmart/haiku-method/haiku-repair/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/gigsmart/haiku-method/haiku-repair"><img src="https://agentmods.dev/badge/skills/gigsmart/haiku-method/haiku-repair.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 | $0.00030 | $0.00682 |
| Opus 5 | $0.00015 | $0.00341 |
| Sonnet 5 | $0.00006 | $0.00136 |
| Haiku 4.5 | $0.00003 | $0.00068 |
Grade A, and why
haiku-repair 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 10d 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 — 41 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Repair
Call haiku_repair for one of the three remaining v4 recovery paths.
When to use this skill
haiku_repair under v4 is narrow. Use it for:
- Drift baseline corruption — the drift sweep reports the baseline can't be read or parsed. The repair tool rebuilds the baseline from the unit's current outputs.
- Worktree relocation — older H·AI·K·U builds (pre-2026-04) rooted
.haiku/worktrees/at the agent's cwd instead of the primary repo root. After upgrading, the misplaced worktrees still exist and confuse the engine. Repair migrates them viagit worktree movefor clean trees; dirty ones are reported for manual resolution. - Mainline PR/MR generation — when an intent branch is already merged into mainline but a fixable issue is found, repair pushes a fix commit and opens a PR against mainline.
When NOT to use this skill
Do not run /haiku:haiku-repair to "fix a wedge" on a v4 intent that won't advance. The fixes it once did under v3 — synthesizing state.json, validating active_stage, enforcing status: completed/active on units — are no-ops on v4 intents. v4 derives stage position from disk (unit FM + branch topology), has no state.json, and uses iterations + approval stamps instead of unit-level status enums. The v0→v4 migrator runs in-band at haiku_run_next time and handles every schema-shape concern.
If a v4 intent won't advance, the right diagnostic is the loop-guard diagnostic in haiku_run_next's response (writes to $TMPDIR/haiku-prompts/{session_id}/loop-guards.log and surfaces a diagnostic: line in the error body). Paste that into a bug report — don't run /haiku:haiku-repair hoping it'll patch something.
Default behavior (git repo)
In a git repository, haiku_repair (no args) scans every intent branch via temp worktrees, applies the narrow fixes above, pushes, and opens PRs when applicable. The loop is sequential because some fixes need user input.
Args
intent: <slug>— repair a single intent in the current working directory only (skips multi-branch mode)apply: false— scan without applying fixes (returns the report only)skip_branches: true— force cwd-only mode even in a git repo
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.
- 10d ago First seen · 41 lines · 30 tokens per session scan A f0e36bfe4136
haiku-repair is a skill published in the GitHub repository gigsmart/haiku-method (24 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 30 tokens to every session and 682 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
git-worktree
A tool for managing Git worktrees, which are separate working folders linked to one Git repository. It supports working on branches in parallel and bringing their changes back to the main branch.
commit-changes
A manually triggered process for preparing and making a precise Git commit or push, where Git is the tool used to track code changes.
deployment-ops
Skill for deployment operations — pre-deployment checklists, deployment pattern selection, rollback procedures, post-deployment verification, and incident response.
automotive-project-management
Comprehensive project planning from initiation to closeout for automotive software projects Covers 2 topics across project-management domain. Includes 2 skill files covering .
parallel-tracks
Run 2-3 features in parallel across independent sessions without breaking the live platform: isolate each feature (branch + git worktree), coordinate through shared repo files (never session memory), and converge via a serialized integration train (rebase -> migration merge -> PR/CI -> ephemeral staging smoke -> merge…
cooperative-vibe-coding
Cooperative vibe coding across machines — real-time collaborative development with humans and AI agents. Covers current tools (Live Share, Code With Me, Zed, Cursor multiplayer, tmux), git-based async patterns, AI agent coordination, and the aspirational future of multi-human multi-agent development. Activate on…