benchmark
Skill1 starsChanged 3 days agoMITView on GitHub
Benchmark and profile the current program (C / C++ first, also Rust and Python) with the tools available (perf stat / record / report, hyperfine, valgrind callgrind / massif, heaptrack, flame graphs), asking before installing a missing tool - optimized build with symbols in a copy, wall time, hardware counters, hot functions (self and inclusive), hot lines, memory - then the analysis of what is slow and why, each finding classified as a quick win or a complex change with its estimated gain, delivered as a PDF / Markdown report (report skill), and optionally the optimizations with a before / after comparison. Use whenever the user asks for a benchmark, a profiling, a performance analysis, what is slow, which function takes the time, or how to optimize a program.
97 other projects use a version of this.
Adopted by
Most recently taken up by
- AndresArreguy/ecc-pi-porttheir own version, first seen Nov 16, 2026
- akshaypimprikar/pragmatheir own version, first seen Nov 11, 2026
- Rojios/everything-claude-codetheir own version, first seen Nov 8, 2026
- accso/ai-context-mcptheir own version, first seen Nov 8, 2026
- timurgaleev/vibestacktheir own version, first seen Nov 8, 2026
- pankaj-mahaur/Agent-Daemontheir own version, first seen Nov 6, 2026
Used with
Projects with it also use ios-clean, ios-fix and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Saqlain-811Skillios-clean
Remove the DebugBridge SPM package and all #if DEBUG wiring from an iOS app. (gstack)
Also in 5 projects that use this
Saqlain-811Skillios-fix
Autonomous iOS bug fixer. (gstack)
Also in 5 projects that use this
seek888Skillpair-agent
Pair a remote AI agent with your browser. One command generates a setup key and prints instructions the other agent can follow to connect. Works with OpenClaw, Hermes, Codex, Cursor, or any agent that can make HTTP requests. The remote agent gets its own tab with scoped access (read+write by default, admin on request). Use when asked to "pair agent", "connect agent", "share browser", "remote browser", "let another agent use my browser", or "give browser access". (gstack) Voice triggers (speech-to-text aliases): "pair agent", "connect agent", "share my browser", "remote browser access".
Also in 11 projects that use this
ShreyRaviSkilllanding-report
Read-only queue dashboard for workspace-aware ship. Shows which VERSION slots are currently claimed by open PRs, which sibling Conductor workspaces have WIP work likely to ship soon, and what slot /ship would pick next. No mutations — just a snapshot. Use when asked to "landing report", "what's in the queue", "show me open PRs", or "which version do I claim next". (gstack)
Also in 8 projects that use this
timurgaleevSkilldocument-generate
Generate missing docs for a feature, module or project as tutorials, how-tos, reference and explanation.
Also in 7 projects that use this
seek888Skillopen-gstack-browser
Launch GStack Browser — AI-controlled Chromium with the sidebar extension baked in. Opens a visible browser window where you can watch every action in real time. The sidebar shows a live activity feed and chat. Anti-bot stealth built in. Use when asked to "open gstack browser", "launch browser", "connect chrome", "open chrome", "real browser", "launch chrome", "side panel", or "control my browser". Voice triggers (speech-to-text aliases): "show me the browser".
Also in 8 projects that use this
- +1 more from Saqlain-811
Other versions97 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 97
- garrytan/gstack★ 142ksame name
- dragonflydb/dragonfly★ 32ksame name
- xtermjs/xterm.js★ 21ksame name
- udecode/plate★ 17ksame name
- Z3Prover/z3★ 13ksame name
- androidx/androidx★ 6.1ksame name
- samchon/typia★ 6ksame name
- PaddlePaddle/FastDeploy★ 3.7ksame name
- zalando/skipper★ 3.3ksame name
- samchon/nestia★ 2.2ksame name
87 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository70 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ask-at-endSkill
- clarifySkill
- continue-workSkill
- exportSkill
- full-permissionSkill
- ideationSkill
- interactive-questionsSkill
- onboardSkill
- pauseSkill
- presentationSkill
- rtkSkill
- setup-contextSkill
- write-like-meSkill
- audit-bugsSkill
- audit-depsSkill
- audit-qualitySkill
- audit-securitySkill
- cmake-styleSkill
- cpp-classSkill
- cpp-commentsSkill
- cpp-projectSkill
- cpp-styleSkill
- better-explain-docSkill
- better-html-docSkill
- better-html-styleSkill
- better-protocol-docSkill
- docs-updateSkill
- explain-docSkill
- html-docSkill
- html-styleSkill
- protocol-docSkill
- readme-styleSkill
- reportSkill
- wiki-styleSkill
- ci-cdSkill
- git-commitSkill
- git-conventionsSkill
- pre-commitSkill
- legal-checkSkill
- legal-setupSkill
- licenseSkill
- libutils-exceptionSkill
- libutils-installSkill
- libutils-setupSkill
- libutilsSkill
- python-classSkill
- python-commentsSkill
- python-projectSkill
- python-styleSkill
- auditSkill
- cppSkill
- devSkill
- docSkill
- gitSkill
- legalSkill
- pythonSkill
- setup-projectSkill
- styleSkill
- setup-studioSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.
0
Common
Most of the 83 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 28
Never install without a yes.
- Distinctive: line 55
read the code, find the cause, estimate the gain (share of the time x expected speedup) and the risk.
- Very specific: line 81
Ask which findings to apply (multiSelect).
“Common” is measured: kept by at least 6 in 10 of the 83 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Benchmark & profiling
SKILL_DIR = the directory of this file. Scripts: scripts/tools.sh (tools available / install commands),
scripts/profile.sh (build + measures), scripts/hotspots.py (tables for the report).
Reports: the report skill. Nothing is committed, the outputs stay in /tmp/benchmark-* unless the user asks.
1. Ask (one AskUserQuestion call, French) - only what the request doesn't say
- Scénario mesuré: the command(s) and the inputs, representative of the real use (a big input, the test suite,
a benchmark target...); propose the obvious one from the project (
./<binary> <typical args>, the GTest binary). - Ce qui compte: temps CPU / latence, mémoire, les deux.
- Profondeur:
Rapide (perf + temps) (Recommandé)·Complète (+ callgrind + mémoire, plus lent). The report format is asked by thereportskill (once).
2. Tools
bash SKILL_DIR/scripts/tools.shLists perf, hyperfine, valgrind, heaptrack, flamegraph, /usr/bin/time, py-spy with the install command of
the OS and the perf_event_paranoid level. When a tool useful for the chosen depth is missing, ask
(AskUserQuestion): Installer (je lance la commande) · Tu les installes (commande donnée) · Continuer sans.
Never install without a yes. Root commands go through the sudo method of the environment (on this computer:
SUDO_ASKPASS=~/.local/bin/sudo-askpass sudo -A <command>, see the global instructions). perf_event_paranoid > 2:
same question for sudo sysctl kernel.perf_event_paranoid=2 (until the next reboot).
Without perf: wall time + callgrind still work; without anything: /usr/bin/time only, say it in the report.
3. Measure
bash SKILL_DIR/scripts/profile.sh --project <root> --cmd "<binary> <args>" [--cwd <inputs dir>] [--runs 10] \
[--callgrind] [--memory] --out /tmp/benchmark-<name>
python3 SKILL_DIR/scripts/hotspots.py /tmp/benchmark-<name> --md /tmp/benchmark-<name>/hotspots.md --json /tmp/benchmark-<name>/hotspots.json- CMake project: rebuilt in a
/tmpcopy (uncommitted changes included) in theOptimizedmode of the user's CMake (Releaseotherwise) with-g -fno-omit-frame-pointer; the copy's binaries come first in thePATHwhile the command runs. Never profile a Debug / Asan build. Already built program:--no-build. - Measures: wall time (hyperfine, else
/usr/bin/timewith the max RSS),perf stat(user space counters: IPC, branch and cache misses),perf record(cpu-clock samples with call graphs: one stream on hybrid CPUs), self / inclusive reports, flame graph (flame.svg) when the FlameGraph scripts exist, callgrind (exact instruction counts, use a small input: 20-50x slower) and heaptrack / massif with--memory. - Variance: stddev > 5 % of the mean -> more
--runs, close the heavy programs, say it in the report. - Rust:
cargo build --releasewith[profile.release] debug = true(orCARGO_PROFILE_RELEASE_DEBUG=true), thenprofile.sh --no-build --cmd "target/release/<bin> ...". Python:py-spy record -o /tmp/py.svg -- python3 x.py(flame graph) andpython3 -m cProfile -s cumtime x.py | head -40.
4. Analyse (the real work: the tables only say where, the code says why)
For every hot spot (top 10 self + the top functions of the program in the inclusive table):
- who calls it:
perf report -i <out>/perf.data --no-children -S '<symbol>' -G --stdio | head -60; - which lines:
perf annotate -i <out>/perf.data '<symbol>' --stdio | head -80(orcallgrind_annotate --auto=yes); - read the code, find the cause, estimate the gain (share of the time x expected speedup) and the risk. Classify each finding:
- Quick win (local change, low risk): regex compiled per call /
std::regexin a loop (hand-written scan, compiled once, or a faster engine), copies by value of big objects, allocations in a hot loop (reserve, reuse buffers), two lookups instead of one (find+[]),std::endlflushes,std::mapwhereunordered_map/ a sorted vector fits,std::function/ virtual calls in the innermost loop, string concatenation in a loop, exceptions as control flow, logging / verbose formatting done even when disabled, I/O without buffering, lock taken per item. - Complex (design change): algorithm complexity (O(n^2) on big inputs), data layout (AoS -> SoA, cache misses),
parallelism (threads,
utils::pool::Cluster), I/O strategy (mmap, batching), caching / memoization, an external library (SIMD, a regex engine). Counters help: IPC < 1 -> memory bound (cache misses, layout); high branch misses -> unpredictable branches; highsystime -> syscalls (strace -c).
5. Report (report skill)
Location audit/<YYYY-MM-DD>-benchmark.md|pdf at the root of the repository. Structure:
- Summary: table
Point | Conclusion(scenario, wall time mean +- stddev, IPC, the 3 hottest spots with their share, the best gains expected). - Measures: environment (
info.txt: CPU, kernel, build type, command), wall time and counters tables (hotspots.md). - Hot spots: the self and the program tables of
hotspots.md, the flame graph (when it exists). - Findings: numbered,
1. **Finding.** cause (file:line), share of the time, fix, expected gain, risk, split in### Quick winsand### Complex changes. - Recommendations: table ordered by gain / effort (
Change | Gain | Effort | Risk). - Reproduce: the exact
profile.shcommand, the output directory.
6. Optimize (only if the user asks, after the report)
Ask which findings to apply (multiSelect). Apply them one by one, keep the behavior (run the tests), re-run the same
profile.sh command (same inputs, same runs) and add a Before | After | Gain table to the report. A change that
doesn't gain more than the variance is reverted and said so.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

