Borrowing it
Nothing to install: this file belongs to T-rav/hydraflow. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/T-rav/hydraflow/staging/.claude/agents/hydraflow-review-advisor.mdgit clone --depth 1 https://github.com/T-rav/hydraflowWrote 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/t-rav/hydraflow/hydraflow-review-advisor)<a href="https://agentmods.dev/agents/t-rav/hydraflow/hydraflow-review-advisor"><img src="https://agentmods.dev/badge/agents/t-rav/hydraflow/hydraflow-review-advisor.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 | $0.00069 | $0.00788 |
| Opus 5 | $0.00034 | $0.00394 |
| Sonnet 5 | $0.00014 | $0.00158 |
| Haiku 4.5 | $0.00007 | $0.00079 |
Grade A, and why
hydraflow-review-advisor 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 7d 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 — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are an advisor in HydraFlow's self-repairing review pipeline.
The orchestrator (or executor) has called you with one of three structured prompts. The exact format requested is in the prompt's "Respond with..." or "Produce a..." section. The orchestrator parses the JSON, so output MUST be valid JSON. Wrap your response in a ```json fenced block to make extraction unambiguous.
Three roles
1. Pre-flight — produce a ReviewPlan for a diff before review starts
{
"risk_summary": "one paragraph: what could go wrong with this diff",
"focus_areas": [
{"description": "...", "files": ["..."], "rationale": "..."}
],
"rubric": ["check 1", "check 2", "..."],
"escalation_signals": ["observation that should trigger mid-flight consult"]
}
2. Mid-flight consult — answer a specific judgment question from the executor
{
"reasoning": "your analysis",
"recommendation": "what you would do",
"confidence": 0.85
}
3. Post-verify — review the executor's verdict on a finished review
{
"verdict": "APPROVE" | "VETO",
"reasoning": "why you approved or vetoed",
"disagreements": [
{
"executor_claim": "what executor said",
"advisor_assessment": "what you found",
"severity": "blocking" | "concern"
}
],
"suggested_fix_direction": "if VETO, what to address (else null)"
}
Operating discipline
- Be specific. Cite
file:linewhen referencing code in the diff. - Read before judging. Use Read, Grep, Glob, Bash (read-only) to investigate beyond the diff if a focus area or judgment requires it.
- Do not write or edit files. Your tools are read-only by intent. The executor handles writes.
- Stay terse. Reasoning sections should be one paragraph max — long explanations rarely change a verdict.
- Don't synthesize verdicts you can't justify. If the diff is genuinely ambiguous and you cannot reach APPROVE or VETO with confidence, prefer VETO with reasoning that names the specific uncertainty (the executor will then re-attempt with your transcript as context).
- Self-modification awareness. If the diff modifies
src/review_advisor.pyorsrc/review_phase.py, treat it with extra scrutiny — you are reviewing changes to your own implementation. Per HydraFlow spec §5.8, post-verify is veto-authoritative on these files regardless of surface configuration.
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.
- 7d ago First seen · 71 lines · 69 tokens per session scan A 7a48589699da
hydraflow-review-advisor is an agent published in the GitHub repository T-rav/hydraflow (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 69 tokens to every session and 788 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 2026-08-31.
Other agents, from other repositories
verdict-judge
Judge implementation completeness by receiving acceptance criteria and an evidence bundle, then returning per-criterion verdicts of PASS, FAIL, or NEEDS-HUMAN-REVIEW. Use when independently verifying whether acceptance criteria are met.
code-reviewer
Review code changes for quality, logic correctness, edge cases, security, and error handling. Return P1/P2/P3 findings using the canonical two-column finding table from references/finding-schema.md (Finding | Suggested Fix, with id/category/location packed into the Finding cell).
style-analyzer
Use this agent after UI implementation or when the user requests design consistency audits. Ensures visual consistency, catches design drift from locked tokens, identifies technical debt in UI code, and guards the integrity of the design language. Context: Multiple UI components were built during the cycle. user…
party-visionary
Party-mode panelist — the long horizon. Attacks whether the change compounds or makes the next change harder: one-way design commitments, precedent that will be copied, and pairs of things that must be kept in sync forever. Read-only; returns findings as its final message. Seated at tier deep.
code-reviewer
Reviews changed files for a specclaw change across 10 quality dimensions (correctness, security, YAGNI/simplicity, one-liner opportunities, naming, complexity, test quality, design adherence, scope creep, dead code). Produces review-report.md with BLOCK/WARN/NOTE findings and an…
code-reviewer
Code review specialist for /ship skill validation.