simiancraft-skills / asset-optimization
AGENTS.md8 starsChanged 4 months agoMITView on GitHub
This skill is built to grow. It ships incomplete by design: the set of asset kinds is stable, but the cookbook of recipes, the tool manifest, and the measured size data all expand as we learn. When you discover a recipe, a tool, or a whole category that is not here yet, encode it and open a PR. An agent that is only using the skill (not editing it) can instead open an issue, with the user's permission, to report…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository23 other files it gives its agents.
- android-emulator-harnessSkill
- android-emulator-mask-testingSkill
- appraise-github-issuesSkill
- asset-optimizationSkill
- burn-down-github-issuesSkill
- carve-github-issueSkill
- code-github-issueSkill
- emoji-readoutSkill
- expo-developer-toolsSkill
- expo-ios-simulatorSkill
- fix-github-issueSkill
- how-to-planSkill
- ios-simulator-flow-evidenceSkill
- ios-simulatorSkill
- ios-simulator-triageSkill
- mobile-accessibilitySkill
- playwright-camera-mask-testingSkill
- playwright-gif-captureSkill
- playwright-harnessSkill
- playwright-storybook-flowsSkill
- prove-github-issueSkill
- prove-work-on-githubSkill
- zone-composerSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 48
If a command has not actually been run in this environment, it is unverified.
- Distinctive: line 50
Keep the result only if it is smaller AND valid.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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
Contributing to asset-optimization
This skill is built to grow. It ships incomplete by design: the set of asset kinds is stable, but the
cookbook of recipes, the tool manifest, and the measured size data all expand as we learn. When you
discover a recipe, a tool, or a whole category that is not here yet, encode it and open a PR. An agent
that is only using the skill (not editing it) can instead open an issue, with the user's
permission, to report an outdated command or a missing recipe or kind; that trigger lives in SKILL.md.
Where new knowledge goes
| You discovered | Add it to |
| a useful command or recipe | the kind's executors/<kind>/cookbook.md |
| a new tool | tools.md (define it once, with its key facts) |
| a real before/after measurement | the kind's executors/<kind>/expectations.md |
| a footgun or caveat | the tool's key-facts row in tools.md, and the recipe's Caveat line |
| a whole new asset kind (rare) | a new executors/<kind>/ folder with the four organs |
Adding a kind or tool: thread every registry
A new kind or tool touches several hand-maintained registries, not just its own file.
A new kind (executors/<kind>/ with the four organs) must also be wired into:
- the dispatch table in
SKILL.md, - the extension table (and, if deferring it, the "Not yet a kind" list) in
classify.md, - the index in
cookbook.md, - its surface mapping in the new
targets.md.
A new tool (a row in tools.md) must also:
- carry a
Kindsvalue that matches which executors actually use it, and - appear in the Prerequisites of every
procedure.mdthat uses it.
The four-organ shape is deliberately reserved even for thin kinds (vector, document, and model each run one or two tools): the fixed slots give a contributor a known place to grow into, so the uniform structure is a decision, not a default.
Cookbook entries
Cookbook recipes follow the entry rubric defined once in cookbook.md: Task, For,
Command, Validate, Gains (marked measured or est, never fabricated), and Caveat, with an optional
See also. When you add a recipe, fill every required field; cookbook.md holds the format and a
worked example.
Honesty (non-negotiable)
- A Gains number is measured or it is labeled
est. Do not record an unmeasured number as fact;expectations.mdholds only observed before/after bytes. - If a command has not actually been run in this environment, it is unverified; say so rather than
implying it was tested. The whole skill is
status: draftuntil its recipes are run and measured. - Keep the result only if it is smaller AND valid; a recipe that can grow a file or break a decode must say so in its Caveat.
Voice
Terse and imperative. No marketing. Prose uses semicolons for independent clauses and the Oxford comma; no em dashes.
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.