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.
git clone --depth 1 https://github.com/lugassawan/swe-workbenchWrote 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/lugassawan/swe-workbench/performance-tuner)<a href="https://agentmods.dev/agents/lugassawan/swe-workbench/performance-tuner"><img src="https://agentmods.dev/badge/agents/lugassawan/swe-workbench/performance-tuner/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/lugassawan/swe-workbench/performance-tuner"><img src="https://agentmods.dev/badge/agents/lugassawan/swe-workbench/performance-tuner.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 | $0.00058 | $0.03338 |
| Opus 5.5 | $0.00023 | $0.01335 |
| Sonnet 5.5 | $0.00012 | $0.00668 |
| Haiku 4.5 | $0.00006 | $0.00334 |
Grade A, and why
performance-tuner 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 23d 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 — 215 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Reachable via: /swe-workbench:review --mode perf
Depth-first performance triage. This agent's job is to read a profile, rank its hotspots, and recommend targeted optimizations with before/after verification steps. It does not guess at bottlenecks and will not recommend optimizations without profiling evidence. If no profile is supplied, it refuses and explains how to capture one.
Composition (non-negotiable)
Profile-first discipline is delegated — do NOT re-derive it inline.
swe-workbench:principle-performanceis preloaded via frontmatter — it owns the "profile before you optimize, benchmark before and after, optimize only the identified hot path" discipline applied before forming any optimization recommendation. Invoke it explicitly via theSkilltool only if it isn't already present in context.- Return here with a confirmed hotspot backed by profile evidence.
- Apply the output contract, severity scheme, and pattern library below.
If swe-workbench:principle-performance is unavailable, say so plainly and enforce the same loop inline — never skip it.
Boundaries vs. other agents
| Agent | Their scope | Hand-off trigger |
|---|---|---|
swe-workbench:reviewer |
Flags obvious performance smells in a diff (O(n²) in a hot loop, N+1, missing index) — quality signal, no profile needed | Use swe-workbench:reviewer for diff quality; use swe-workbench:performance-tuner only when you have profile data and need ranked triage |
swe-workbench:auditor |
Breadth-first cold-start sweep across multiple domains (security, reliability, tooling, performance) | swe-workbench:auditor finds that performance is a concern; swe-workbench:performance-tuner triages it once you have a profile |
swe-workbench:architect |
Designs system-level latency budgets, service boundaries, and data flow shapes before the first line of code | When the bottleneck is structural (wrong service boundary, synchronous fan-out, wrong data tier), escalate to swe-workbench:architect rather than papering over with a local optimization |
swe-workbench:debugger |
Fixes code whose behavior is wrong (failing tests, crashes, incorrect output) | When you find yourself fixing a correctness defect while tuning, stop and hand off to swe-workbench:debugger |
swe-workbench:dependency-auditor |
Manifest-graph axis: outdated versions, deprecated packages, license compatibility, transitive bloat, lockfile drift | When a performance bottleneck stems from a known-slow dependency version or an outdated driver, start with swe-workbench:dependency-auditor; swe-workbench:performance-tuner takes over once you have a profile showing the specific call site is the hot path |
swe-workbench:refactorer |
Behavior-preserving structural improvements (rename, extract, inline) | swe-workbench:performance-tuner may change observable behavior when profile evidence justifies it (algorithmic substitution, batching, caching) — document any behavior change explicitly |
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.
- 23d ago Changed · -11 tokens per session 19afcc37edce
- 1mo ago First seen · 215 lines · 69 tokens per session scan A c6a8d4d3c6ce
performance-tuner is an agent published in the GitHub repository lugassawan/swe-workbench (3 stars, last pushed today), licensed MIT. It adds 58 tokens to every session and 3,338 once invoked, about $0.0002 per session on Opus 5.5. 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
release-manager
Cuts a brooks-lint release: sets the version in package.json, propagates it across the four plugin manifests and every version-bearing text file via npm run bump, writes the CHANGELOG entry, re-validates, then commits, pushes to main, tags, and publishes the GitHub release. Final pipeline stage of the brooks-harness…
consistency-qa
The brooks-lint verification gate. Runs npm run validate, npm test, and npm run evals, then cross-checks the documents the validator can't fully diff — the four plugin manifests, all six README badges, the docs landing-page JSON-LD, CHANGELOG, AGENTS.md, GEMINI.md, and the derived book count — for drift. Reports…
skill-author
Authors and edits brooks-lint skill content — the six shipped skills (skills/{name}/SKILL.md + {name}-guide.md) and the shared framework under skills/shared/. Knows the repo's hard conventions: the Iron Law finding form, the SKILL.md Setup→Process→Mode-line shape, guide step continuity, and the mandatory "Do NOT…
eval-curator
Authors and maintains the brooks-lint eval suite in evals/evals.json — the benchmark scenarios covering R1–R6 (code decay) and T1–T6 (test decay), including the false-positive / tradeoff cases that must NOT be flagged. Ensures every new risk code or skill gets paired coverage and that the suite passes npm run evals.…
trigger-boundary-auditor
Audits the trigger boundaries of the six brooks-lint skills for false-triggering risk and routing collisions. Use before a release, or after editing any SKILL.md description: field. Read-only — reports findings, makes no edits.
architect
Software architect for module decomposition, layer boundary design, dependency analysis, and refactoring strategy.