Imported from howie1329/Howard-AgentKit (
templates/AGENTS.md). Install upstream withnpx skills add howie1329/Howard-AgentKit --skill templates. Copyright stays with the author.
[Project Name] Agent Guide
Purpose
This is the primary instruction file for AI coding agents working in this repository.
[Project Name] is [short project description]. Agents should optimize for safe, focused, production-ready changes that follow local patterns.
Source Of Truth
Follow this file first. It defines the repository-wide operating rules for agents.
Use companion guides only when a trigger below applies. Do not load or restate every guide by default.
For routine bugfixes and small features, do not open WORKFLOWS.md, planning templates, or optional guides unless a trigger applies or the user asks.
When To Use Other Guides
| Trigger | Action |
|---|---|
| Any file edits (final message) | Use the change-explanation format in After Coding (Required); see CHANGE-EXPLANATION.md for detail |
| UI, components, styling, layout | Read DESIGN.md or [design system path, e.g. docs/design-system.md] when set |
| Review, refactor, dependencies | Read CODE-QUALITY.md |
| Tests, fixtures, QA strategy | Read TESTING.md if present |
| Auth, permissions, secrets, PII | Read SECURITY-CHECKLIST.md if present |
| Stack-specific code | Read STACK.md if present |
| User or issue requests a PRD | Create from PRD-TEMPLATE.md under [briefs path, e.g. docs/briefs] |
| User or issue requests an implementation brief | Create from IMPLEMENTATION-BRIEF-TEMPLATE.md under [briefs path, e.g. docs/briefs] |
| Release or branching process (optional) | Read WORKFLOWS.md if present |
Optional Guides (Full Template Set)
These files are not installed with the standard template set. Do not assume they exist unless present in the repository:
TESTING.mdSECURITY-CHECKLIST.mdPRD-TEMPLATE.mdIMPLEMENTATION-BRIEF-TEMPLATE.mdWORKFLOWS.md
Planning Artifacts
Use this decision tree before creating planning documents:
- Tiny obvious fix — implement directly; no PRD or brief required.
- Product uncertainty — create a PRD from
PRD-TEMPLATE.mdonly when the user, issue, or task explicitly requests planning documentation. Save it under[briefs path, e.g. docs/briefs]. - Multi-file engineering with meaningful risk — create an implementation brief from
IMPLEMENTATION-BRIEF-TEMPLATE.mdonly when the user, issue, or task explicitly requests it. Save it under[briefs path, e.g. docs/briefs]. - Security-sensitive change — follow
SECURITY-CHECKLIST.mdif present.
Do not read planning templates on every task. Copy and fill them only when planning documentation is requested.
Local Guidance
This root AGENTS.md applies to the whole repository.
If a subdirectory contains its own AGENTS.md or equivalent local guidance, follow the most specific local guidance for files in that subtree. When instructions conflict, prefer the more specific local guidance unless it violates repository-wide safety, security, or quality rules.
Operating Rules
- Inspect existing code and patterns before changing files.
- Prefer the smallest complete solution that satisfies the task.
- Keep diffs focused and reviewable.
- Preserve user changes. Do not overwrite or revert work you did not make.
- Avoid speculative abstractions, broad rewrites, and dependency bloat.
- Validate external input at system boundaries.
- Keep public APIs and persisted data shapes stable unless the task requires a change.
- Ask before destructive commands, broad refactors, schema changes, dependency additions, or security-sensitive changes.
- Run the narrowest useful checks before finishing (use Project Commands below).
- After any coding work that changes files, end with a change explanation per After Coding (Required). Do not skip this for small fixes.
- Watch for common AI mistakes: invented APIs or commands, placeholder logic that looks production-ready, client-only authorization, and tests that do not exercise changed behavior.
- When reviewing or flagging issues, use severity: Blocker (must fix), Concern (should fix soon), Suggestion (optional).
- For review, refactor, or dependency work, read
CODE-QUALITY.mdfor the full checklist.
Before Coding
Before making meaningful code changes:
- Confirm the task, scope, and acceptance criteria.
- Inspect nearby code, tests, and existing patterns.
- Read at most one triggered companion from When To Use Other Guides if it applies.
- For non-trivial work: link to or create an issue in [issue tracker, e.g. Linear or GitHub Issues] and use an appropriate branch when the project expects branches.
- Ask for clarification if requirements, risk, or approval boundaries are unclear.
Tiny fixes can skip issue-and-branch ceremony when the change is obvious and low risk.
During Coding
While implementing:
- Modify existing patterns before creating new ones.
- Keep behavior explicit and easy to test.
- Avoid unrelated formatting churn.
- Prefer existing utilities, components, and conventions.
- Add or update tests when behavior changes.
- For UI work, follow
[design system path, e.g. docs/design-system.md]and preserve existing interaction patterns. - For stack-specific work, read
STACK.mdwhen present. - Document meaningful decisions in the change explanation when they matter.
Approval Boundaries
Ask before:
- Adding a new runtime dependency.
- Changing schemas, migrations, permissions, auth, billing, or data retention behavior.
- Running destructive commands or deleting files.
- Changing global design tokens, theme primitives, or foundational layout rules.
- Performing broad refactors or large formatting-only rewrites.
- Editing generated, vendored, or external-source files unless the task explicitly requires it.
After Coding (Required)
If you modified any files in the repository, your final message must include a change explanation.
This is how developers catch up on agent work. It is required for bugfixes, refactors, and small edits—not only large features or PRs. Scale section depth to the change; do not omit the explanation because the diff is small.
Do not write a separate change-explanation file unless the project asks for one. Use this structure in your reply:
- Summary — outcome in plain language (behavior delivered, not only files touched).
- What changed — main files or areas; what, why, and how each fits existing patterns.
- Key decisions — approach, alternatives, trade-offs (skip when straightforward).
- Verification — checks run and results; note any skipped checks and why.
- Risks or limitations — merge blockers, untested paths, follow-up (use "None identified" for trivial low-risk fixes when appropriate).
- Suggested review focus — where the developer should look first.
See CHANGE-EXPLANATION.md for examples and style rules.
Before Finishing
Before marking work complete:
- Re-read the original request and acceptance criteria.
- Confirm the implementation satisfies the requested scope.
- Run the narrowest relevant checks from Project Commands.
- Review the diff for unrelated changes.
- Confirm no unnecessary dependencies, schema changes, theme changes, or broad refactors were introduced.
- Include the change explanation from After Coding (Required) in your final message.
Project Commands
Replace these examples with the project's real commands:
| Command | Description |
|---|---|
npm run dev |
Start the local development server |
npm test |
Run tests |
npm run lint |
Run lint checks |
npm run build |
Build the project |
Remove commands that do not apply.
Environment
Document required environment variables here:
[ENV_VAR_NAME]: [description]
Never commit API keys, tokens, private keys, or local .env values. Keep secrets in environment variables or the project-approved secret store.
Stack
When STACK.md is present (for example after a preset install), read it before changing stack-specific code.
Otherwise document the real stack for this project:
- [Primary framework]
- [Language/runtime]
- [Backend/data layer]
- [Styling system]
- [Test tools]
- [Lint/format tools]