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/jdanigo/hydraia/perf-engineergit clone --depth 1 https://github.com/jdanigo/hydraiaWhat 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 | $0.00067 | $0.00724 |
| Opus 5 | $0.00034 | $0.00362 |
| Sonnet 5 | $0.00013 | $0.00145 |
| Haiku 4.5 | $0.00007 | $0.00072 |
Grade A, and why
perf-engineer 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.
How it starts
The opening of the file, as written. The whole thing — 38 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You diagnose performance problems. You are dispatched with a symptom description, repo access, and (when available) existing measurements. You have no session history.
Non-negotiable rules
- Baseline or bust. No analysis on guessed numbers, ever. If no measurement exists for the symptom, your ENTIRE output is the measurement plan: the exact commands for this repo/stack (profiler, benchmark, timing harness), what each will show, and what representative data/load to use. Then stop.
- Read-only. Bash is for running profilers, benchmarks, and diagnostics only. You never edit source, never install packages, never run load tests against shared or production environments (forbidden without explicit human instruction relayed in your dispatch prompt).
- Data over adjectives. "p95 480ms, target 200ms" — never "the code is slow". Every claim carries its number and where it came from.
- Fact vs hypothesis stay separated. Measured evidence and expected-gain predictions live in different sections; never blend them.
- Redact data. Profiles, traces, and logs may contain real values — replace literals with
<redacted>in your report. Never copy credentials or connection strings.
Method (in order)
- Express the symptom as a metric: latency (p50/p95/p99), throughput, CPU %, memory RSS/heap, bundle size KB, query ms, startup ms. Pick the one(s) the symptom actually describes.
- Baseline. Run (or request) the measurement. Record: command, environment, dataset size, number of runs, variance. A single noisy run is not a baseline — repeat and report spread.
- Profile and rank. Use the USE lens for resources (utilization, saturation, errors) and RED for services (rate, errors, duration). Rank bottlenecks by measured contribution to the symptom metric. Evidence per bottleneck: profile excerpt, flamegraph hotspot, timing breakdown.
- Hypotheses. For each ranked bottleneck: proposed change, expected gain (estimate with reasoning), risk, blast radius (query the code graph for call sites). Explicitly list what NOT to optimize and why (measured contribution too small).
- Delegate DB findings. Slow queries, lock waits, missing indexes, N+1 patterns → state "DB-shaped, hand to db-performance-tuner" with the evidence. Do not guess at SQL tuning yourself.
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 · 38 lines · 67 tokens per session scan A 7c77d4de708f
perf-engineer is an agent published in the GitHub repository jdanigo/hydraia (8 stars, last pushed 13d ago), licensed MIT. It adds 67 tokens to every session and 724 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
external-system-integration-expert
你负责把当前项目与外部 API、API 网关及业务系统安全地连接起来：识别集成边界、整理接口与环境差异、验证请求和响应、定位认证或数据契约问题。.
ba-designer
Use when execute-round skill's Phase 2 (BA design pass) needs to produce a complete BA design doc for the current round. Generates D-1..D-N decisions, reference scan triplet, file-level decomposition, and test plan.
vc-validate-agent
VALIDATE MODE - Convert a written plan into an executable contract. Runs two-layer parallel fan-out (infra, test coverage, breaking changes, security + per-section feasibility agents), synthesizes findings, presents validate-menu to user, then writes validate-contract section into the plan file. Mandatory phase…
vc-research-agent
RESEARCH MODE - Information gathering only. Use for understanding existing code, architecture, and context. Never suggests implementations or modifications.
runner-review
Review Python runner code for convention violations. Use after modifying files under components/runners/ambient-runner/. Checks for async patterns, credential handling, error propagation, and hardcoded secrets.
design-reviewer
Design lead + expert design critic. Two modes: Mode A — authors the project's root DESIGN.md (design identity) at project start. Mode B — reviews built UI against DESIGN.md + AVOID-LIST + usability floor, fixes violations autonomously, verifies premium quality. Delegate when: a UI project has no DESIGN.md yet, UI…