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/bouob/agent-harness/harness-engineeringnpx skills add bouob/agent-harness --skill harness-engineeringgit clone --depth 1 https://github.com/bouob/agent-harnessWrote 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/bouob/agent-harness/harness-engineering)<a href="https://agentmods.dev/skills/bouob/agent-harness/harness-engineering"><img src="https://agentmods.dev/badge/skills/bouob/agent-harness/harness-engineering.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 | $0.00167 | $0.02312 |
| Opus 5 | $0.00084 | $0.01156 |
| Sonnet 5 | $0.00033 | $0.00462 |
| Haiku 4.5 | $0.00017 | $0.00231 |
Grade A, and why
harness-engineering 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.
How it starts
The opening of the file, as written. The whole thing — 182 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/harness-engineering — Multi-Agent Harness Framework
A flexible framework for the Planner-Generator-Evaluator (P-G-E) design pattern as
formalized by Anthropic Engineering on 2026-04-04. Where /sprint is a fixed 6-phase
pipeline producing .sprint/<ts>/ artifacts, this skill helps you reason about,
construct, or review harnesses that don't fit a rigid template — including the
harness elements (skills/rules/hooks/references) that surround them.
Core Premise
Every component in a harness encodes an assumption about what the model can't do on its own — and those assumptions are worth stress-testing. (Anthropic, 2026-04-04)
The harness is durable goods; the model is a replaceable engine. As models strengthen, remove non-load-bearing scaffolding. As failure modes appear, add elements that survive model swaps (Hooks > Skills > Rules > References > Prompts on the durability axis).
Input
$ARGUMENTS
If empty: ask whether the user wants (a) execution of a P-G-E task, (b) design review of an existing skill/rule/hook, (c) routing/decomposition advice, or (d) diagnosis of a failed harness run — then proceed.
Procedure
Step 1 — Classify the Request
State the chosen mode in one line before proceeding. Do not collapse modes — they have different downstream procedures.
| Mode | Trigger | Goes to |
|---|---|---|
| Execute | "do X using multi-agent / subagents / P-G-E" | Step 2 |
| Design | "should this be a hook / skill / rule?" or "review this skill" | Step 6 |
| Route | "which model should handle X?" or "spawn subagent or inline?" | Step 5 |
| Diagnose | "why did the harness fail?" / context drift / stale handoff | Step 7 |
Step 2 — Plan: Expand the Spec
Apply Planner discipline. The Planner's job is to expand a brief prompt (1–4 sentences) into a detailed spec — not to write code. Output:
- Re-stated goal (one sentence)
- Deliverables (files, decisions, reports)
- Sprint contract: testable success criteria with hard thresholds
- Decomposition into 2–7 tasks; each:
- Self-contained (cold-start agent can execute with task text alone)
- Measurable (acceptance criteria start with a verb: Returns, Renders, Stores)
- Scoped (≤50k tokens of work)
- Dependencies → tasks with no deps go to
parallel_batch; with deps go tosequential_tasks.
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.
- 4d ago First seen · 182 lines · 167 tokens per session scan A 10b4038b9414
harness-engineering is a skill published in the GitHub repository bouob/agent-harness (5 stars, last pushed 1mo ago), licensed MIT. It adds 167 tokens to every session and 2,312 once invoked, about $0.0008 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
ci-review
CI-optimized code review: multi-agent parallel review with confidence scoring and atomic GitHub PR review posting. Runs specialized review agents, scores findings, and submits a GitHub PR review via post-review.sh. Always executes post-review.sh to submit a review on every run (including zero-findings runs). Triggers…
pr-check
PR review compliance: fetch review comments from an open PR, categorize as resolved/unresolved/dismissed, critically evaluate fixable items, implement approved fixes, and reply inline. Pass --unattended to halt on human-judgment items (emitted as Pending-Human) instead of prompting via AskUserQuestion.
council
Consult external AI council (Gemini 3.8 Flash, Codex, GLM-5.3, Kimi) for thorough reviews and consensus-driven decisions. Use ONLY when explicitly invoked with "/council" or when user says "consult the council", "invoke council", or "council review". Do NOT auto-trigger on generic phrases like "thorough review".
babysit
PR babysitter: monitors CI status, auto-rebases when behind, auto-fixes CI where possible, delegates review comment handling to dlc:pr-check, and re-requests review after fixes. Designed for /loop usage with Remote Control.
pr-validity
PR validity analysis: fetch PR diff, extract new code additions, search existing codebase for duplicate or overlapping implementations, classify changes, and create a structured GitHub issue.
brainstorm
You MUST use this skill before /pm when work is ambiguous, scope is unclear, or multiple approaches exist. Explores problem space, proposes 2-3 approaches with trade-offs, and produces an approved spec document that feeds into both issue creation (/pm) and implementation planning (/cdt). Triggers: brainstorm, explore…