constitution-lite
Skill6 starsChanged 3 months agoMITView on GitHub
Define a tiny set of stable, non-negotiable agent authority boundaries for repeated or delegated work without duplicating CLAUDE.md, AGENTS.md, or task specs.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- analyze-miniSkill
- bug-captureSkill
- build-oneSkill
- ceremony-budgetSkill
- checklist-miniSkill
- context-checkSkill
- diagnose-loopSkill
- eco-modeSkill
- falsify-specSkill
- grill-with-docs-liteSkill
- handoffSkill
- lean-modeSkill
- mini-specSkill
- operator-profileSkill
- scope-freezeSkill
- ship-miniSkill
- teach-backSkill
- test-miniSkill
- thin-planSkill
- tool-noise-guardSkill
- verify-contractSkill
- workspace-checkpointSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 8 are distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Use when repeated, delegated, scheduled, or autonomous work needs a small set of stable constraints that should survive across tasks, such …
- Distinctive: line 41
Read existing project instructions first.
- Distinctive: line 43
Keep a candidate only when violating it should require a human decision, explicit review, or a hard stop across more than one task.
- Distinctive: line 46
Remove architecture preferences, commands, style guidance, framework choices, and task-local constraints.
- Distinctive: line 47
If the constitution conflicts with live project instructions or a task contract, stop and ask for a human decision.
- Distinctive: line 49
Keep the artifact short enough to scan before delegated work.
- Distinctive: line 58
Use the constitution to narrow delegated authority, not to expand it.
- Distinctive: line 82
Every retained rule changes agent authority across multiple tasks.
≈ “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
Constitution Lite
Purpose
Record only the durable rules that constrain what an agent may decide or do across multiple tasks. This is an authority boundary, not a project handbook.
Project instructions are not constitution rules. Architecture notes, repo layout,
commands, coding style, naming conventions, framework choices, and ordinary team
preferences belong in native project instructions such as CLAUDE.md or AGENTS.md.
Task-specific requirements belong in SPEC.md, scope-freeze, or SHIP.md.
When to use
Use when repeated, delegated, scheduled, or autonomous work needs a small set of stable constraints that should survive across tasks, such as:
- human approval before destructive or external side effects
- protected production, data, credential, or customer boundaries
- tests or evidence that an agent may not weaken to make work pass
- decisions an agent is never authorized to make alone
Skip this skill when the need is only project onboarding, commands, architecture, style, a temporary implementation constraint, or one task's acceptance criteria.
Inputs
- Existing
CLAUDE.md,AGENTS.md, or equivalent project instructions if present - Stable risk and authority boundaries
- Protected resources, environments, data, or interfaces
- Human approval requirements
- Existing task or shipping controls that should not be duplicated
Workflow
- Read existing project instructions first. Do not copy their conventions into the constitution.
- List candidate cross-task rules.
- Keep a candidate only when violating it should require a human decision, explicit review, or a hard stop across more than one task.
- Classify each retained rule as
MUST,MUST NOT, orHUMAN GATE. - Make each rule observable where possible: name the protected resource, forbidden action, approval event, or evidence requirement.
- Remove architecture preferences, commands, style guidance, framework choices, and task-local constraints.
- If the constitution conflicts with live project instructions or a task contract, stop and ask for a human decision. Do not silently choose an authority winner.
- Name who may amend the constitution and how the amendment is approved.
- Keep the artifact short enough to scan before delegated work.
Authority boundary
CONSTITUTION.md is a workflow artifact. It does not override system instructions,
organization policy, platform permissions, sandboxing, or runtime safety controls.
It also does not grant an agent permission merely because an action is absent from
the constitution.
Use the constitution to narrow delegated authority, not to expand it.
Outputs
CONSTITUTION.md- Non-negotiable
MUST/MUST NOTinvariants - Explicit
HUMAN GATEdecisions - Protected resource or data boundaries when relevant
- Amendment owner and process
Canonical shape:
CONSTITUTION
MUST: ...
MUST NOT: ...
HUMAN GATE: ...
Protected boundary: ...
Conflict rule: stop and ask for a human decision
Amendment owner: ...Stop conditions
- Every retained rule changes agent authority across multiple tasks.
- Ordinary project instructions and task-local constraints have been removed.
- Human approval gates are explicit.
- A conflict has been surfaced rather than silently resolved.
- The artifact is small enough to scan before delegated work.
Anti-patterns
- Rewriting
CLAUDE.mdorAGENTS.mdasCONSTITUTION.md. - Filling the artifact with architecture, commands, coding style, or naming conventions.
- Promoting temporary preferences into permanent governance.
- Treating the constitution as permission to perform otherwise unauthorized actions.
- Adding aspirational principles that cannot change an agent decision.
- Silently resolving conflicts between durable governance and a task contract.
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.

