write-agent-prompts
Skill22 starsChanged 39 days agoMITView on GitHub
Write prompts and skills for capable peer agents using high-salience technical referents, conceptual breadth, and semantic compression. Use when shaping how an agent recognizes task structure, recruits knowledge, selects methods, and adapts its approach.
In this repository2 other files it gives its agents.
- proptest-praxisSkill
- write-plansSkill
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 28
Use references/concepts.md when a referent needs further grounding or its scope matters.
≈ “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
Write Agent Prompts
Write for a capable peer. Load as many distinct, relevant, high-salience referents as possible, including adjacent methods and ways of seeing. Use common ground to compress exposition and make room for breadth; use structural relationships to make that repertoire composable. This skill demonstrates the technique on prompt construction itself; reproduce the technique in the target domain.
Common ground; audience design; pragmatic compression. Use the recipient's shared technical vocabulary. An established concept can carry a family of methods, assumptions, examples, and failure modes. Spend explanation on local distinctions that change its application. Ground ambiguous terms; expand unfamiliar acronyms. Prefer recognized terminology and clear relationships to invented labels or cryptic shorthand.
Cognitive task analysis; recognition-primed decision making; affordances. Recover the expertise the prompt should convey: what makes a situation interesting, which cues are diagnostic, what an expert expects next, and what changes the choice of approach. Extract this from the user's examples, corrections, and priorities. Preserve the underlying distinctions. Describe the opportunities and constraints the recipient should learn to recognize, including ones that have no obvious keyword match.
Problem representation; framing; representational change. Consider what the task becomes under different representations: a search problem, a transformation, a coordination problem, an inference, a design tradeoff. Choose entities and relationships that expose useful structure: dependencies, state transitions, invariants, information flow, incentives, audiences, or asymmetries between doing and checking. Supply grounds for changing the frame when it hides important features or stops explaining observations.
Associative retrieval; conceptual coverage; divergent and convergent search. Survey core methods, neighboring disciplines, alternative formulations, representations, explanatory lenses, and characteristic failures. Include plausible adjacencies before a specific use is known: a concept can make a future opportunity recognizable. Search for unfamiliar connections as well as resolving uncertainty about familiar ones; use primary sources to ground unfamiliar methods. Organize the repertoire by meaningful relationships—complements, alternatives, prerequisites, tensions, reductions—so concepts can be combined. Trim repetition while preserving distinct useful associations.
Structural alignment; systematicity; analogical transfer. Transfer connected relationships across domains. Identify what makes an analogy productive and where the correspondence fails. Use examples and contrast cases to expose those relationships; vary surface vocabulary while preserving the relevant structure. A target catalog supplies starting points for extrapolation. A method name may stand alone when its meaning is shared; add application detail where it changes recognition, selection, or execution.
Rational metareasoning; value of computation; exploration/exploitation. Allocate research and writing effort to conceptual gaps that could change the recipient's approach. Distinguish the organizing frame, available repertoire, and actual requirements. Where the task benefits from adaptive work, convey useful tradeoffs and grounds for continuing, switching methods, or stopping. Leave ordinary technical choices to the recipient. Recover goals, deliverables, capabilities, and constraints from context; clarify missing information when it materially changes the guidance.
Partial evaluation; staging; abstraction boundaries. Specialize this writing method to the domain, recipient, and objective. The resulting prompt should carry the relevant domain concepts, concrete context, and consequential distinctions. Keep authoring instructions at the authoring level; include metalevel concepts in the result when they help the target task. Let the task determine the structure. A repertoire is available knowledge, not a requirement to execute every named method.
Modeling; worked examples; self-application. Demonstrate the technique in the guidance itself. Let precise referents, complementary lenses, structural relationships, and selective grounding show the recipient how to construct similar guidance. The demonstration should make the selection and use of concepts legible. Reproduce the generative pattern across new domains; choose the vocabulary each domain warrants.
Construct validity; transfer; counterfactual evaluation. Read the prompt as its recipient: what becomes noticeable, which hypotheses or methods become available, and what decisions could change? Try a representative case, a structurally similar case with different vocabulary, and a case where the framing would fail. Assess the prompt by the behavior and artifacts it supports. When running evaluations, compare downstream outcomes and account for interactions between cues. Distinguish observed performance from a predicted effect of the wording.
Use references/concepts.md when a referent needs further grounding or its scope matters. These are lenses for constructing guidance; select and combine them as the task warrants.
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.

