tool-discipline
Skill1 starsChanged 59 days agoMITView on GitHub
Shared discipline for available tools, complete evidence and safe delivery.
In this repository42 other files it gives its agents.
- AGENTS.mdAGENTS.md
- content-safetySkill
- context-compactionSkill
- coordinator-communicationSkill
- cortex-controlSkill
- design-frontend-designSkill
- design-impeccableSkill
- design-reference-analysisSkill
- design-ui-ux-pro-maxSkill
- design-visual-qaSkill
- design-workflowSkill
- documentation-syncSkill
- knowledge-harvest-scopeSkill
- knowledge-harvestSkill
- orchestratorSkill
- requirements-intakeSkill
- skill-authoringSkill
- skill-managementSkill
- tool-discipline-workerSkill
- worker-accessibility-auditorSkill
- worker-accessibility-fixerSkill
- worker-architectSkill
- worker-backend-devSkill
- worker-build-verificationSkill
- worker-code-reviewerSkill
- worker-database-architectSkill
- worker-data-engineerSkill
- worker-debuggerSkill
- worker-devops-engineerSkill
- worker-explorerSkill
- worker-frontend-devSkill
- worker-fullstack-devSkill
- worker-generalSkill
- worker-mobile-devSkill
- worker-performance-engineerSkill
- worker-plannerSkill
- worker-qa-engineerSkill
- worker-refactorerSkill
- worker-security-auditorSkill
- worker-senior-consultantSkill
- worker-technical-writerSkill
- worker-ux-designerSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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 8
Before a call, identify the concrete fact or state change it must resolve.
- Distinctive: line 15
Read each tool's complete live description and schema before first use and after context loss, including tools reached through a search …
- Distinctive: line 23
Follow the advertised arguments, bounds, preconditions and retry contract.
- Distinctive: line 32
Use real same-attempt check receipts.
- Distinctive: line 51
Read current targets before changing them and preserve unrelated work.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Tool discipline
Before a call, identify the concrete fact or state change it must resolve. Reuse retained results while relevant state is unchanged. A timeout, quiet worker or desire to reconfirm does not justify another call.
Runtime owns all delegation; native subagents are disabled. Use only tools available to this host and role. Cortex MCP owns lifecycle operations, reports, questions and skill reads; external MCP servers supply their domain tools. Read each tool's complete live description and schema before first use and after context loss, including tools reached through a search layer; names and recovery summaries are insufficient. In code mode, take Cortex argument shapes from the schema-derived tool cheat sheet, and fetch one exact record with tools_discover only when a field's bounds or conditions matter or a call was refused. Otherwise, when offered, use tools_discover rather than guessing from a names-only tool list. It does not discover external MCP schemas or grant access. For a listed skill, read its full listed description, not only its name. Preflight required and conditional fields against the live schema or cheat sheet before calling. Measure bounded strings in its counting unit; rewrite overlong text without silently truncating requirements or weakening verification. Follow the advertised arguments, bounds, preconditions and retry contract; never guess synonyms, probe errors to discover requirements, inspect private runtime state, or copy argument contracts into skills.
Batch independent reads; keep dependent reads, mutations and approvals sequential. When offered, read_batch groups eligible independent Cortex evidence pages using its live schema; it is not an evaluator or a transaction. Preserve each item's status and native paging metadata. Deferred, failed or unrun items remain incomplete until resolved; an outer reply alone proves no inner success. Inspect success, errors and truncation before relying on a result. A wrapper must expose the complete nested result, including exit status or running handle. Inspect content if structuredContent is absent. Use real same-attempt check receipts. Continue an owned running handle to a terminal result; never launch an equivalent command while it remains active. Keep waits bounded. If it cannot finish, close only that handle and report its timeout or cancellation as unverified. A truncated result is incomplete evidence: resolve the missing fact. Observer-only rendering truncation does not itself prove a native call failed.
Retain acknowledged identifiers; never replay a mutation for reassurance. Correct one deterministic error only when the live declaration makes the fix unambiguous; otherwise stop that route. An access, channel, role, scope or policy refusal is a decision, not a glitch: never reach the same effect another way. Request the needed authorization through the runtime or report the blocker.
Recognized missing, incomplete or stale file-read preconditions are not access refusals: nothing was changed. If your role edits files, read the current target on this channel, then rebuild and re-propose the edit under the existing authority, without an owner question or extra approval. This never cures a real denial or permits switching tools, channels or workers to bypass one.
Read current targets before changing them and preserve unrelated work. Drafts use the owned gateway; recover draft identities through bounded public reads, not guessed paths. Record failures, unrun checks and missing evidence without secrets, private report bodies or raw host logs.
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.

