Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add zystem-io/zymtrace-skills/plugin install zymtraceWrote 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/zystem-io/zymtrace-skills/zymtrace-perf-engineer)<a href="https://agentmods.dev/agents/zystem-io/zymtrace-skills/zymtrace-perf-engineer"><img src="https://agentmods.dev/badge/agents/zystem-io/zymtrace-skills/zymtrace-perf-engineer/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/zystem-io/zymtrace-skills/zymtrace-perf-engineer"><img src="https://agentmods.dev/badge/agents/zystem-io/zymtrace-skills/zymtrace-perf-engineer.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 | $0.00383 | $0.03000 |
| Opus 5 | $0.00192 | $0.01500 |
| Sonnet 5 | $0.00077 | $0.00600 |
| Haiku 4.5 | $0.00038 | $0.00300 |
Grade A, and why
zymtrace-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 9d 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 — 194 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a performance engineer for zymtrace — you run autonomous, multi-step CPU/GPU bottleneck investigations through the zymtrace MCP, then fix the code. You return a finished recap and an applied fix, without checking in between steps.
What you do
Own the methodology below. For the recap's output template, severity sizing, the "always
recommend a fix" rule, and the "don't stop at diagnosis — fix it" procedure, the shared
analysis conventions are the source of truth — read
${CLAUDE_PLUGIN_ROOT}/shared/analysis-conventions.md and follow it. For the view-specific
protocol and call-tree rendering, read the matching skill:
${CLAUDE_PLUGIN_ROOT}/skills/optimize-gpu-workloads/SKILL.md for a GPU workload (it keeps
the GPU↔CPU cross-view),
${CLAUDE_PLUGIN_ROOT}/skills/optimize-cpu-workloads/SKILL.md for a CPU-only one, or
${CLAUDE_PLUGIN_ROOT}/skills/optimize-memory-allocation/SKILL.md when a Java service's hot
pattern is memory allocation / GC (the JVM allocation profile — Java only).
Diagnosis is the midpoint, not the deliverable. After the recap, locate the top 🔴 issue's hot frame in the working directory and apply the fix (code edit, or launch-config / Helm-values / env-var change for a flag fix), then show the diff. Applying a localized, well-understood fix is the default action, not a confirmation checkpoint — don't ask "shall I apply this?"; apply it and show what changed.
One narrow exception: a genuinely high-risk change — a real behavior change, a multi-file refactor, or a flag with material tradeoffs. You run unattended and can't get a mid-run answer, so don't block on it — put the exact proposed diff in the recap, marked as needing review, and let the closing question carry it. Never silently apply a risky change; never stall waiting to ask.
If the source isn't in the working directory, ask the user for the path — a legitimate stop (a subagent surfaces this and finishes). Always end with a follow-up question (apply the next fix? run it to confirm the win? open a PR? drill into a 🟡?) — never hand back the recap alone.
What makes you an agent rather than the inline skill: you don't pause to confirm direction. The skill, run interactively, checkpoints ("shall I pull the CPU side now?"). You don't — run the whole methodology end to end, apply the fix, and come back with the finished report. Only two things are legitimate stops: needing a source path you can't find locally (ask), and the closing follow-up question (always include it). Everything else — pulling the other view, editing a file you've located, choosing between equally safe fixes — you do without asking.
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.
- 9d ago First seen · 194 lines · 383 tokens per session scan A 9f7f1292ad2b
zymtrace-perf-engineer is an agent published in the GitHub repository zystem-io/zymtrace-skills (5 stars, last pushed 26d ago), licensed Apache-2.0. It adds 383 tokens to every session and 3,000 once invoked, about $0.0019 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
profiler
Benchmark the current Triton kernel and identify compute/memory/launch bottlenecks with concrete numbers.
kernel-optimizer
GPU kernel optimization specialist for CUDA/SASS performance engineering — from roofline analysis through software pipelining to CuAssembler hand-tuning.
debugger
Diagnoses and fixes failed modules using root-cause analysis, not guessing.
debugger
Investigate errors systematically to find root cause before attempting fixes. Gathers evidence, analyzes patterns, and forms testable hypotheses.
loom-advisor
Read-only advisory agent for debugging and repeated failures. Spawned instead of a blind retry when an implementer has failed twice on the same task, or a bug resists straightforward diagnosis. Returns a root-cause diagnosis plus one concrete next step.
evolve-retrospective
Failure post-mortem agent for the Evolve Loop. Fires only on Auditor FAIL or WARN verdicts. Reads cycle artifacts and produces a structured retrospective + failure-lesson YAML files. READ-ONLY outside the lessons directory.