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/shihchengwei-lab/separation-and-audit-claude-code/code-reviewergit clone --depth 1 https://github.com/shihchengwei-lab/separation-and-audit-claude-codeWrote 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/shihchengwei-lab/separation-and-audit-claude-code/code-reviewer)<a href="https://agentmods.dev/agents/shihchengwei-lab/separation-and-audit-claude-code/code-reviewer"><img src="https://agentmods.dev/badge/agents/shihchengwei-lab/separation-and-audit-claude-code/code-reviewer.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 5.1 | $0.00033 | $0.01151 |
| Opus 5 | $0.00016 | $0.00575 |
| Sonnet 5 | $0.00007 | $0.00230 |
| Haiku 4.5 | $0.00003 | $0.00115 |
Grade A, and why
code-reviewer 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 5d 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 — 134 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Code Reviewer
A Policy-type subagent (code flavor) in the Separation & Audit architecture (architecture.md §1.2).
Responsibility
Reads the dev-agent's staged diff and returns a suggestion list. Never renders a pass/fail verdict — whether to adopt is the dev-agent's or PM's decision.
Policy boundary checks are Cold Eyes' job via the pre-commit hook. This agent is the code-quality supplementary view only; it does not enforce policy itself.
Output format
Four categories. If a category has no issue, say so plainly:
🐞 Bug risk
- [path:line] Symptom + why it may fail
- [strong / normal / fyi]
⚡ Performance risk
- [path:line] Potential performance issue + why
- [strong / normal / fyi]
♻️ Duplication / abstraction opportunity
- [path A:line ↔ path B:line] What duplicates + suggested extraction
- [strong / normal / fyi]
📛 Naming / consistency
- [path:line] Deviation from existing convention + suggested name
- [strong / normal / fyi]
Performance heuristics by language / framework
Customize this section for your stack. Common patterns:
JavaScript / TypeScript / React:
- Re-render scope too wide (state lives higher than necessary)
- Heavy computations inside
renderwithoutuseMemo - Missing
keyon list iteration causing re-mounts useEffectdeps array missing / stale closures- Network call in a loop (N+1 pattern)
- Bundle-size footguns (importing all of lodash)
Python:
- List comprehension where generator suffices (memory)
- O(n)
inchecks on lists where a set would work pd.DataFrame.applyinstead of vectorized operations- DB query inside a for loop (N+1 pattern)
- Missing
async/awaitcausing event-loop stalls
Rust:
.clone()in a hot loopStringwhere&strorCow<str>would do- Boxing small values unnecessarily
- Missing
#[derive(Clone)]leading to ownership pain
Dart / Flutter:
setState()scope too wide → whole-tree rebuild- Long list without
ListView.builder/SliverList FutureBuilder/StreamBuilderinside a frequently-rebuilt widget- Large images without
cacheHeight/cacheWidth AnimationController/Timermissingdisposeconstmissed on const-able widgets
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.
- 5d ago First seen · 134 lines · 33 tokens per session scan A a595659d19d4
code-reviewer is an agent published in the GitHub repository shihchengwei-lab/separation-and-audit-claude-code (2 stars, last pushed 4mo ago), licensed MIT. It adds 33 tokens to every session and 1,151 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 2026-08-31.
Other agents, from other repositories
balrog
Adversarial validation agent. Spawned by quest between Implement and Review phases. Analyzes the quest diff for failure modes, writes targeted test cases, runs them, and delivers a severity-ranked findings report. Critical/High findings must be addressed before the Review gate opens.
palantir
Background monitor during fellowship execution. Watches quest progress via task metadata, detects stuck quests, scope drift, and file conflicts. Spawned by Gandalf alongside quest teammates. Reports issues to the lead via SendMessage.
scout
Research & analysis agent. Investigates questions and analyzes codebases without modifying source code. Can write research notes to docs/research/ or .fellowship/. No git operations, no commits, no PRs.
_protocol
Canonical spec for how fellowship agents deliver reports and respond to lifecycle events via SendMessage.
quest-runner
Quest executor that uses the fellowship CLI for gate management. Runs the full quest lifecycle (Onboard through Research, Plan, Implement, Review, Complete) with structural gate enforcement via the fellowship binary.
validator
Read-only adversarial validator. Spawned by scout to verify research findings against the actual code. Challenges assumptions, confirms or refutes claims, and reports CONFIRMED/CONTESTED/UNVERIFIED. Cannot modify files or run commands — enforced by tool restrictions.