design-agent-capability
Skill1 starsChanged 4 months agoMITView on GitHub
Design or consolidate durable agent capabilities: instructions, skills, scripts, integrations, hooks, workflows, subagents, or bundles. Use for repeated workflows or overlap; inventory existing capabilities first.
In this repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- apply-project-updatesSkill
- audit-project-updatesSkill
- engineer-cli-toolsSkill
- maintain-agent-instructionsSkill
- maintain-project-docsSkill
- maintain-project-memorySkill
- maintain-web-design-systemSkill
- prepare-web-assetsSkill
- red-team-technical-planSkill
- release-web-applicationSkill
- research-software-versionsSkill
- verify-software-changeSkill
- web-discovery-artifactsSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 8 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
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
- Very specific: line 41
Do not assume every client implements subagents, hooks, plugins, or workflows in the same way.
- Very specific: line 47
Use the open Agent Skills core: one directory, exact SKILL.md, and YAML name plus description.
- Distinctive: line 49
Put all trigger information in the description.
- Distinctive: line 50
Keep the body imperative, focused, and below 500 lines.
- Distinctive: line 52
Add scripts only for deterministic work that is repeatedly reimplemented.
- Distinctive: line 53
Use relative paths within the skill.
- Distinctive: line 56
Keep exact hosts, paths, credentials, device IDs, repositories, and release commands in repository-local overlays.
- Distinctive: line 58
Add optional client metadata only in namespaced locations that other hosts can ignore safely.
- Distinctive: line 79
Test explicit invocation and representative implicit triggers.
- Distinctive: line 80
Test a near-miss prompt that should not trigger.
≈ “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
Design Agent Capability
Choose the smallest durable artifact that solves a demonstrated repeated problem.
This skill chooses and designs the capability type. Once durable instruction
files are selected, use maintain-agent-instructions to audit or edit their
actual client and repository surfaces.
Inventory first
Inspect applicable:
- global and repository instructions
- installed and repository-local skills
- scripts, CLIs, hooks, and CI enforcement
- MCP servers and app connectors
- client-specific commands, workflows, and subagents
- runbooks and templates
Identify ownership and overlap. Prefer strengthening the current owner over creating another broadly triggered capability.
Choose the artifact
| Need | Preferred artifact |
| One task or temporary constraint | Prompt/thread context |
| Always-on repository convention | Repository instruction |
| Reusable on-demand procedure or expertise | Agent Skill |
| Deterministic repeatable transformation | Script or CLI |
| External live data or authorized action | MCP/app connector |
| Mechanical lifecycle enforcement | Hook or CI |
| Scheduled or chained operation | Client workflow/automation |
| Isolated delegated role | Client-specific subagent |
| Multi-capability distribution | Client plugin/bundle |
Do not assume every client implements subagents, hooks, plugins, or workflows in
the same way. Read references/client-surfaces.md when cross-client support is
required.
Design a portable skill
- Use the open Agent Skills core: one directory, exact
SKILL.md, and YAMLnameplusdescription. - Put all trigger information in the description.
- Keep the body imperative, focused, and below 500 lines.
- Move conditional detail into directly linked
references/. - Add scripts only for deterministic work that is repeatedly reimplemented.
- Use relative paths within the skill.
- Describe capabilities before client-specific tool names; provide adapters and fallbacks when a tool is optional.
- Keep exact hosts, paths, credentials, device IDs, repositories, and release commands in repository-local overlays.
- Add optional client metadata only in namespaced locations that other hosts can ignore safely.
Define the contract
Specify:
- discriminative triggers and non-triggers
- required inputs and discovered context
- workflow and decision gates
- allowed mutations and approval boundaries
- tools, optional dependencies, and fallbacks
- failure behavior
- evidence and output contract
- maintenance owner and update source
Validate
- Run structural validation.
- Scan for personal paths, private hosts, credentials, and client-only assumptions.
- Test explicit invocation and representative implicit triggers.
- Test a near-miss prompt that should not trigger.
- Forward-test on a synthetic repository when behavior is complex.
- Confirm instructions do not claim precedence a host does not provide.
Return the inventory, artifact decision, structure, safety boundaries, portability assessment, tests, and installation options.
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.