workflow-optimizer
Skill0 starsChanged 54 days agoMITView on GitHub
Make a recurring job cheaper. Measure a baseline, test cheaper variants against the same quality check, keep the winner as a skill. Use when the user wants a repeat workflow (reports, content batches, triage, builds) to cost less ("optimize this workflow", "get this under 1% of my usage").
4 other projects use a version of this.
Adopted by
Most recently taken up by
- iberi22/skillstheir own version, first seen Oct 15, 2026
- poorvith-mp/skills-developertheir own version, first seen Aug 19, 2026
- roborracle/agent-workflow-kittheir own version, first seen Aug 11, 2026
- Shanvision-star/paper-aigc-content-factorytheir own version, first seen Jun 20, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- Shanvision-star/paper-aigc-content-factory★ 2same name
- iberi22/skills★ 0same name
- poorvith-mp/skills-developer★ 0same name
- roborracle/agent-workflow-kit★ 0same name
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 repository10 other files it gives its agents.
- cheap-subagentsSkill
- context-auditSkill
- limit-window-warmupSkill
- pinpointSkill
- progress-ledgerSkill
- quiet-toolsSkill
- short-answersSkill
- skillifySkill
- token-saverSkill
- usage-budgetSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
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
- Distinctive: line 21
Write a pass/fail quality check: tests pass, output matches the format, a reviewer checklist.
- Distinctive: line 25
Run the job once as it is today and record the cost with whatever meter the host has:
- Distinctive: line 55
Keep a change only if the check still passes.
- Distinctive: line 56
Stop when the experiment budget is spent or two ideas in a row save next to nothing.
- Very specific: line 60
Write the winning procedure as a skill (see skillify), with scripts for the deterministic parts and the quality check as its final step.
- Very specific: line 68
No host documents giving the model its own plan usage today (see usage-budget), so the model can't check a target like that by itself.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Workflow optimizer
A one-off task isn't worth optimizing. A job you run every day or week is: spend some usage once to find the cheap way, and every run after that costs less. Treat cost like any other bug. Measure, change one thing, measure again, with a quality check that stops "cheaper" from quietly meaning "worse".
0. Is it worth it?
Experiments cost roughly (baseline cost) × (number of test runs). If the job won't run enough times to pay that back, stop here and say so.
1. Pin the job and the quality bar
- Fix one set of test inputs and reuse it for every run.
- Write a pass/fail quality check: tests pass, output matches the format, a reviewer checklist. Without one, no saving can be trusted.
2. Measure a baseline
Run the job once as it is today and record the cost with whatever meter the host has:
| Host | Meter |
| Claude Code | /usage before and after (plan share); per-session tokens from the local logs, e.g. npx ccusage@latest session |
| Codex | /status before and after; npx ccusage@latest codex daily |
| Gemini CLI | /stats model; npx ccusage@latest gemini daily |
| API | the usage fields on each response |
Plan meters move in coarse steps, so a small job can show no change at all; use token counts where you can. Also note tool calls, files read and wall time. They explain where the cost went.
3. Candidate changes, roughly cheapest first
- Hand it the inputs (paths, templates, IDs) instead of letting it search. See the
pinpointskill. - Move deterministic steps into a script: renaming, packaging, formatting, API calls with fixed parameters.
- Batch several items per run when they share context, instead of one run per item.
- Send mechanical steps to a cheaper model and keep review on the strong one. See
cheap-subagents. - Quiet the tools and shorten the replies. See
quiet-toolsandshort-answers. - Switch off tools, plugins and MCP servers the job never uses. See
context-audit. - Lower the reasoning effort for the mechanical parts only.
4. Experiment
- One change (or one small bundle) per run, same inputs, same quality check.
- Log every run:
| # | Change | Cost | vs baseline | Quality check | Notes |
| 0 | baseline | pass |
- Keep a change only if the check still passes. Drop changes that save almost nothing: complexity has a cost too.
- Stop when the experiment budget is spent or two ideas in a row save next to nothing.
5. Lock it in
Write the winning procedure as a skill (see skillify), with scripts for the deterministic parts and the quality check as its final step. Record the measured cost per run inside it, so drift shows up later.
6. Report
Baseline vs optimized cost per unit, how many runs, what changed, what didn't help. Say plainly that small samples on one input set are indicative, not proof, and re-check on real inputs.
About "get this under 1% of my limit"
No host documents giving the model its own plan usage today (see usage-budget), so the model can't check a target like that by itself. State the target anyway, so it aims for it, and measure between runs yourself as above.
Before you finish
- Baseline and every variant measured with the same meter and the same inputs.
- The quality check passed for the version that was kept.
- The winner is saved as a skill, or the reason it wasn't is stated.
Part of agent-skills · MIT · improvements welcome as issues.
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.