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 manhua-man/harness-coding-protocol --skill harness-initgit clone --depth 1 https://github.com/manhua-man/harness-coding-protocolWrote 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/manhua-man/harness-coding-protocol/harness-init)<a href="https://agentmods.dev/skills/manhua-man/harness-coding-protocol/harness-init"><img src="https://agentmods.dev/badge/skills/manhua-man/harness-coding-protocol/harness-init/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/manhua-man/harness-coding-protocol/harness-init"><img src="https://agentmods.dev/badge/skills/manhua-man/harness-coding-protocol/harness-init.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 352 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00028 | $0.07371 |
| Opus 5.5 | $0.00011 | $0.02948 |
| Sonnet 5.5 | $0.00006 | $0.01474 |
| Haiku 4.5 | $0.00003 | $0.00737 |
Grade A, and why
harness-init 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 yesterday.
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 — 430 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Harness Init — agent-native
You are the IDE_Agent. /harness-init runs in five phases against the user's target repository:
Phase 1 — Ground → Phase 2 — Read → Phase 3 — Judge & Write Drafts → Phase 4 — Show & Confirm → Phase 5 — Apply
All five phases are agent-native. Do not require Node.js, npm, npx, tsx, a package manager, or a project-local runtime. You read the project, judge what each Root_Truth_File should say, produce Drafts in working memory, ask for one confirmation, and then write only the confirmed root entry and resolved feature documentation targets. Nothing lands on disk until the user types yes in Phase 4.
If the user supplied a project path argument, use it. Otherwise, use the current working directory.
Targets (Root_Truth_Files and conditional feature documents)
The allowed targets below are agent-written outputs (no template-only bytes):
-
<target>/AGENTS.md— project facts -
<target>/CLAUDE.md— collaboration protocol -
<target>/DESIGN.md— product, visual, interaction, brand, content, or DX design system; conditional, see DESIGN gate below -
<target>/steering/harness-recommendations.md— local steering overrides; conditional, see steering gate below -
Optional Cursor adapter (one logical Root_Truth_File, two physical paths) — only when
.cursor/is already present or the user asks for Cursor rules:<target>/.cursor/rules/harness.mdc<target>/.cursor/commands/harness-init.md
Slash commands in target repos: this plugin ships
ai-ide/commands/andai-ide/skills/. In a target repo, the user chooses where IDE slash commands live (ai-ide/commands/,commands/,.cursor/commands/,.claude/commands/, etc.). Do not scaffold duplicate command trees across paths unless the user requests it. Reference the chosen path fromAGENTS.md; skip the Cursor pair when.cursor/is absent and the user keeps commands elsewhere. -
Conditional feature documentation — when existing feature docs, implementation evidence, or explicit requirements support a useful catalog: the resolved documentation index, named feature pages, and an existing status summary only when affected. Read the feature documentation contract before selecting paths. For a tiny project, prefer a section in its existing README; for an empty project without requirements, skip with a reason. Each actual path must appear separately in the Phase 4 summary. This is not authorization to write arbitrary
docs/**.
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- yesterday Changed · +3 lines 0653a1204cab
- 5d ago Changed · +3 lines dcb0f77d292a
- 1mo ago Changed · +8 lines · -10 tokens per session 5ca1647327d5
- 1mo ago Changed 196ca957751a
- 1mo ago Changed · +30 lines · -15 tokens per session b16777d2918a
- 1mo ago First seen · 386 lines · 53 tokens per session scan A 86f897858a95
harness-init is a skill published in the GitHub repository manhua-man/harness-coding-protocol (5 stars, last pushed yesterday), licensed MIT. It adds 28 tokens to every session and 7,371 once invoked, about $0.0001 per session on Opus 5.5. 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
investigate
Manage structured investigations with multi-agent evidence collection, synthesis, tagging, and PHI sanitization. Use when user says "start an investigation", "root cause analysis", or "collect evidence about X". Subcommands: new, run, collect, synthesize, close, status, list, search. Do NOT use for casual debugging or…
playbook
Analyze your Claude Code configuration and suggest improvements based on the agentic coding playbook. Use when user says "audit my config", "check my setup", or "improve my CLAUDE.md". Works on both global and project-level CLAUDE.md files. Do NOT use for application configuration — only for Claude Code setup.
checkpoint
Save all work, update memory, commit, push, and prepare to end the session. Use when user says "save my work", "wrap up", or "I'm done for now". Use at natural breakpoints or when context is getting large.
create-project
Scaffold a new project with git, .gitignore, CLAUDE.md, AGENTS.md, GitHub repo, and memory directory. Use when user says "start a new project", "create a repo", or "scaffold an app". Creates the project as a sibling to the .claude/ config directory.
learn
Capture a lesson as a structured knowledge entry. Use when you discover a non-obvious bug, workaround, or pattern worth preserving for future sessions. Use when user says "remember this", "save this lesson", or "this is worth noting".
promote
Promote a project-level lesson to global scope so it's available across all projects. Use when user says "make this global", "share this across projects", or "promote this lesson". Use when a lesson discovered in one project would benefit others.