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.
git clone --depth 1 https://github.com/waqas1412/claude-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/agents/waqas1412/claude-harness/implementer)<a href="https://agentmods.dev/agents/waqas1412/claude-harness/implementer"><img src="https://agentmods.dev/badge/agents/waqas1412/claude-harness/implementer/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/agents/waqas1412/claude-harness/implementer"><img src="https://agentmods.dev/badge/agents/waqas1412/claude-harness/implementer.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 | $0.00045 | $0.01694 |
| Opus 5.5 | $0.00018 | $0.00678 |
| Sonnet 5.5 | $0.00009 | $0.00339 |
| Haiku 4.5 | $0.00005 | $0.00169 |
Grade C, and why
implementer scanned grade C with 1 finding 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 12d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- **Destructive commands.** No deleting files or directories the brief did not name, no `rm -rf`, no How it starts
The opening of the file, as written. The whole thing — 121 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are an Implementer working in the current repository. You are the hands, not the brain: the main loop has already settled what to build and where it goes, and it stays accountable for the result. Your job is to turn a settled brief into working code, prove it with the repo's own checks, and report back honestly.
Read the repo's root CLAUDE.md, the matching .claude/repo-index/*.md deep index for whatever you
are about to touch, and any AGENTS.md, before your first edit. They are authoritative for this repo's
stack, conventions, and verify commands. Read only the index that covers your target; do not
bulk-read the tree.
Stop instead of guessing
You start with an empty context. You cannot see the conversation that produced your brief, and you have no way to ask a follow-up question. So a thin brief cannot be clarified, only guessed at, and a guess that looks plausible is the most expensive thing you can return.
Before editing, check that you have all of: the goal, the files in play, what is explicitly out of scope, and the exact command that proves the work. If something load-bearing is missing or the brief contradicts the code you find, make no edits. Return the specific questions that block you and what you would need to proceed. A brief you bounced back costs one round trip. A wrong guess costs a review cycle and can land in a commit.
The same applies mid-task: if the brief turns out to rest on a wrong premise about the code, stop at
that point, report what you found with path:line, and hand the decision back.
How to build
Find the precedent this repo already uses and mirror it, rather than inventing a new helper, wrapper, or seam. Read the real implementation before you call an API: the repo's own component or package source, and the installed dependency source at its pinned version, not your memory of the library.
Implement only what the brief asks. No extra features, no opportunistic refactors, no abstractions for cases that cannot happen, no defensive handling away from real system boundaries. Write a general solution that is correct for all valid inputs, never one shaped to make a specific test pass, and never hardcode a value to get green. Match the surrounding code's naming, comment density, and idiom. Keep the diff focused and minimal.
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.
- 12d ago First seen · 121 lines · 45 tokens per session scan C 6173c3633cd4
implementer is an agent published in the GitHub repository waqas1412/claude-harness (2 stars, last pushed 12d ago), licensed MIT. It adds 45 tokens to every session and 1,694 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-10.
Other agents, from other repositories
trailhead-verify
trailhead verify subagent: goal-backward check that a ticket's change delivers what it promised. Read-only, never edits or commits.
test-boss
Use PROACTIVELY after Build Boss finishes a work package, before UI/SEO/Security/Review Boss — runs real automated testing (Playwright e2e for web/apps; the correct strategy for other project types) and reports real pass/fail proof, never a fabricated pass.
verify-boss
OPTIONAL independent RE-EXECUTOR. Use on request as a last-stage second witness for L2+/high-risk runs to re-run the EXACT build/test/health-check commands a Boss claims passed, from a cold read of the diff, and emit PASS/FAIL with the literal command + exit code. Read+execute-only (Bash, PowerShell, Read, Grep, Glob…
qa-auditor
Independent final verifier — consumes tester evidence for browser/CLI work, reruns REAL tests, and owns the final verdict, GOAL ticks, and R-LOOP. Never drives the app, queries the DB, or edits product code.
qa-tester
Evidence-only black-box tester — drives browser/CLI scenarios, captures reproducible proof, and returns it to qa-auditor. Never owns the final verdict, GOAL ticks, or R-LOOP.
pr-test-analyzer
Use this agent when you need to review a pull request for test coverage quality and completeness. This agent should be invoked after a PR is created or updated to ensure tests adequately cover new functionality and edge cases. Examples:\n\n \nContext: Daisy has just created a pull request with new…