skill-awareness
Skill14 starsChanged 8 months agoMITView on GitHub
Use when starting any session — maps contexts to behavioral skills that should be active
In this repository50 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adr-conventionSkill
- architecture-vocabularySkill
- backlog-externalSkill
- backlog-localSkill
- backlogSkill
- checkpointsSkill
- design-principlesSkill
- domain-glossarySkill
- git-practicesSkill
- grounded-claimsSkill
- knowledge-checkSkill
- policy-riskSkill
- positive-framingSkill
- receiving-code-reviewSkill
- skill-authoringSkill
- subagent-driven-developmentSkill
- test-driven-developmentSkill
- token-efficientSkill
- triageSkill
- verification-before-completionSkill
- workflow-bugSkill
- workflow-checkSkill
- workflow-commitSkill
- workflow-contractsSkill
- workflow-debugSkill
- workflow-decisionsSkill
- workflow-docsSkill
- workflow-doctorSkill
- workflow-epicSkill
- workflow-featureSkill
- workflow-flow-fix-pipelineSkill
- workflow-flowSkill
- workflow-grillSkill
- workflow-handoffSkill
- workflow-ideaSkill
- workflow-implementSkill
- workflow-planSkill
- workflow-proposalSkill
- workflow-prSkill
- workflow-refineSkill
- workflow-researchSkill
- workflow-reviewSkill
- workflow-startSkill
- workflow-statusSkill
- workflow-tech-reviewSkill
- workflow-update-workflowSkill
- workflow-validateSkill
- workflow-worktreeSkill
- workflow-yoloSkill
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 17
Verify it if it's load-bearing: If a code change, plan step, command, or founder decision would rest on the claim, confirm it *before* …
≈ “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
Skill Awareness
Behavioral skills must be active when relevant, even in ad-hoc sessions without slash commands. This mapping tells you which skills to load based on what you're doing.
Always Active — Grounded Claims
This rule needs no trigger. It applies from the first message of every session, including ad-hoc ones.
A recalled fact and a verified fact must never look the same in your output.
- Label it. Any claim about this repo, an external library, or a memory/prior-session fact that you have not confirmed with a tool call in this session is marked inline:
(unverified — from training),(unverified — prior session). A verified claim citesfile:line, a quoted snippet, or command output instead. - Verify it if it's load-bearing. If a code change, plan step, command, or founder decision would rest on the claim, confirm it before building on it — however confident you are. If you can't tell whether it's load-bearing, it is.
- Gaps stay open. "I don't know, and here's what would tell us" is a complete answer. Never close a gap with a plausible guess.
Load virtual-team:grounded-claims for the full protocol and the rationalization catalog.
Context-to-Skill Mapping
| Context | Skill to load | Trigger |
| Starting a pipeline | virtual-team:triage | Before /flow executes any step, or when /feature//implement need to determine ceremony level |
| Writing production code | virtual-team:test-driven-development | Before any Edit/Write to non-test files (reads stack.md tdd: field for mode: strict, recommended, off) |
| Writing production code | virtual-team:design-principles | Before writing function signatures, constructors, or service boundaries (reads stack.md design: field for mode: strict, recommended, off) |
| Claiming completion | virtual-team:verification-before-completion | Before saying "done", "complete", "passes" |
| Asserting a fact a decision rests on | virtual-team:grounded-claims | Before stating how existing code, a dependency, or a recalled fact behaves (summary above is always active) |
| Defining or changing behavior around payments, app-store distribution, login, personal data, or third-party APIs | virtual-team:policy-risk | Before committing to a design that a store, payment provider, or regulator could reject |
| Receiving review feedback | virtual-team:receiving-code-review | When processing review comments |
Executing multi-task plan with --sdd | virtual-team:subagent-driven-development | When /virtual-team:implement --sdd is active |
| User requests compressed output | virtual-team:token-efficient | User says "caveman mode", "terse", "compress", or similar activation phrase |
| Generating client-facing prose in /idea or /proposal | virtual-team:positive-framing | When /idea Round 5 or /proposal Phase 6 reaches the editorial step |
Project-provided domain and stack skills are NOT auto-triggered. They are discovered and loaded by /virtual-team:implement Layer 1 based on domain and stack frontmatter fields matching the current work.
Integration
- Loaded by:
SessionStarthook inhooks/hooks.json - Reinforced by:
PreToolUsehook onEdit|Writecalls (TDD + verification check) - Does NOT replace: command-level skill loading in
/virtual-team:implement(Layer 0, 1, 2)
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.