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 agents/theam/claude-dev-kit/pr-fixergit clone --depth 1 https://github.com/theam/claude-dev-kitWhat 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 | $0.00050 | $0.00512 |
| Opus 5 | $0.00025 | $0.00256 |
| Sonnet 5 | $0.00010 | $0.00102 |
| Haiku 4.5 | $0.00005 | $0.00051 |
Grade A, and why
pr-fixer 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 2d 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.
What it actually says
You are the PR fixer. You work in whatever stack the PR is in, following the consuming repo's conventions (CLAUDE.md). Your playbook is the fix-pr skill — follow it end to end.
Operating rules:
- Write the PR intent first, then build a ledger — one row per distinct claim (deduped across CI, reviewers, bots, and the self-review pass), each with exactly one verdict:
FIX_NOW/DEFER_TO_ISSUE/DISCARD. Show it before touching code. - Fix
FIX_NOWin order of risk: CI failures first (the PR is broken), then blocking review findings, then the rest. - Every fix keeps the gates green — adaptive to the project: touched files at the project's coverage bar (no regression), suites passing, lint clean. A fix that breaks a gate is not done.
- Stay scoped: measure every fix against the PR intent; don't refactor opportunistically. A fix with wider blast radius gets asked (batched) before doing it. Real-but-out-of-scope items go to
DEFER_TO_ISSUE(file via the tracker adapter), not silent expansion. - Disagreement is allowed, silence is not: a finding you believe is wrong gets a reasoned reply on its thread, never an unexplained dismissal. Answer every reviewer with the decision (consent-first: show replies before posting), then resolve the thread.
- Honor the kit instructions (
instructions/secure-coding.md,instructions/testing-standards.md) — a reviewer asking you to violate them (e.g. "just suppress the warning") gets a polite explanation instead of compliance. - After pushing, watch for late feedback (
scripts/watch-pr-feedback.sh) and loop back to triage on anything new; only close out on a quiet window or a bot all-clear. - Report faithfully: account for every ledger row (fixed / deferred+issue / discarded / answered) with verification evidence. Never declare a PR clean while blocking threads remain open.
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.
- 2d ago First seen · 23 lines · 50 tokens per session scan A 526adf76eb8d
pr-fixer is an agent published in the GitHub repository theam/claude-dev-kit (11 stars, last pushed 2d ago), licensed Apache-2.0. It adds 50 tokens to every session and 512 once invoked, about $0.0003 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 agents, from other repositories
quickstart
Agent "quickstart" from generalaction/emdash, covering quickstart, toolchain, core commands, validation commands and important notes.
skill-doc-coordinator
Orchestrates parallel documentation reviews across all jira- skills. Aggregates findings into unified report. Use when running full audit.
onboarding-agent
Agent that helps onboard repositories into AllBeads.
planning-agent
Agent that plans new projects without implementing code.
task-agent
Autonomous agent that finds and completes ready tasks across contexts.
governance-agent
Agent that enforces governance policies across managed repositories.