efficient-md
Skill3 starsChanged 32 days agoMITView on GitHub
Shapes a markdown artifact's length and structure by how long it stays loaded and who reads it, agent or person. Use when writing or restructuring an AGENTS.md or README, a skill description or body, a memory index, a brief, spec, or blueprint, a state or handover file, a decision ledger, or a dispatch's output format. Not for reviewing or rendering a verdict on text that already exists — do that directly, no skill.
In this repository8 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- releaseSkill
- advisor-modeSkill
- memory-refineSkill
- mine-sessionSkill
- plain-englishSkill
- product-engineeringSkill
- section-by-sectionSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
Not a class: where a person is the primary audience, readability governs form.
≈ “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
One axis prices what a multi-agent system authors for a context: how long the tokens stay resident, and in whose context. Name the class first; template, budget, and prohibitions follow. Full doctrine and evidence: references/model.md.
| Class | Residency | Author with |
| HOT | loaded every session by standing machinery (AGENTS.md, memory index, frontmatter descriptions) | references/hot.md |
| WARM | loaded per dispatch or reference (skill bodies, briefs, specs, research tables, living state files) | references/warm.md |
| COLD | write-once, verbatim, never rewritten (filed rulings, post-mortems, transcripts) | references/cold.md |
| CHANNEL | resident in the orchestrator all session (dispatch prompts, returns) | references/channel.md |
Human Audience Overlay
Not a class: where a person is the primary audience, readability governs form; the class still prices substance — file the bulk, present the digest. A surface whose purpose is agent loading never takes the overlay; a dual-read artifact (a README agents also consult) stays overlaid per precedence (2). Where the overlay applies, agent-reader laws — no narrative history, IDs over prose — are SUSPENDED, not merely softened: a human doc's narrative can be its load-bearing substance.
Precedence
(1) shipped law and documented invariants beat class laws; (2) dual-audience artifacts split surfaces before blending laws — never cut a human doc to a WARM ceiling; (3) an artifact may carry a NAMED CONTRACT that replaces class-generic law — a contract lives beside the law it extends and ships only where that law ships.
Procedure
- In scope? The unit must enter a context — class attaches to the unit, never the whole file.
- Name the class by who loads it, how often, for how long; overlay if a person is the primary audience; named-contract artifacts follow their contract.
- Load the class's reference; author under its template and laws.
- On class change, re-price: inline content becomes a pointer.
- SHIP RULE (a ship-time step, the mover's, not the author's): a shipped file may cite only what ships with it — replace unconditional citations of session-scoped or absent artifacts with self-contained statements or in-repo pointers; a citation guarded by an existence check is the sanctioned fallback, not a violation. Any compact rendering keeps the carve-out or drops the whole step.
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.