Agent skill
psi
Plan-spec-implement workflow for structured development. Only use when explicitly directed by user or when mentioned in project AGENTS.md file. Generates ephemeral plans in ~/.dot-agent/, applies specs to project docs, then implements test-first.
Install this agent skill to your Project
npx add-skill https://github.com/siviter-xyz/dot-agent/tree/main/skills/psi
SKILL.md
PSI - Plan Spec Implement
Structured workflow for planning, specifying, and implementing changes with documentation-first approach.
When to Use
Only use when:
- Explicitly directed by the user
- Mentioned in project
AGENTS.mdfile
Do not use automatically - this is an opt-in workflow, not a default.
Core Workflow
Plan → Spec → Implement
Phases are independent - you can start with any phase, but all must ensure documentation stays up-to-date.
Key Principles
- Ephemeral planning - Plans stored in
~/.dot-agent/(not committed) - Documentation-first - Specs applied to project docs/READMEs/AGENTS.md
- Test-first implementation - Tests for docs/user journeys before code
- Design/review embedded - Design and review integrated into Plan/Spec phases
- Phase independence - Each phase can work standalone, all update docs
Phase Overview
Plan Phase
- Generates detailed plans in
~/.dot-agent/repo/YYYY-MM-work-name.plan.md - Research stored in
~/.dot-agent/working-dir/repo/YYYY-MM-work-name.research.md - Embeds design considerations
- Includes review before proceeding
Spec Phase
- Generates specs for: API schemas, interfaces, DTOs, database models, config, env vars, architecture, user journeys, package structure, tech choices
- Embeds design considerations
- Reviews specs before applying
- Applies to:
docs/, README.md files, AGENTS.md files
Implement Phase
- Test-first: tests for docs/user journeys before code
- CI verification: verify types, tests, lint pass before committing
- Atomic commits: group related changes with tests
- Updates docs, README.md, AGENTS.md as code evolves
- Can work independently if specs exist in docs
Research Management
Detects phrases like:
- "looking at your research" → Loads research file
- "refine your research" → Updates research file, narrows focus
Documentation Structure
- README.md: Aim for < 1000 lines (not hard rule), can be longer if needed
- AGENTS.md: < 200 lines, inline at root/packages/modules/code level
- docs/: Architecture, roadmap, tech-choices, setup/, user-journeys/, design/
References
For detailed protocols, see:
references/plan-phase.md- Plan generation with embedded design/reviewreferences/spec-phase.md- Spec generation and applicationreferences/implement-phase.md- Test-first implementationreferences/review-protocol.md- Reviewing plans/specs/designreferences/research-management.md- Research file handlingreferences/docs-structure.md- Documentation organization rulesreferences/file-paths.md- Storage paths and conventions
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
frontend-engineer
Frontend development guidelines for React/TypeScript applications. Modern patterns including Suspense, lazy loading, useSuspenseQuery, file organization with features directory, MUI v7 styling, TanStack Router, performance optimization, and TypeScript best practices. Use when creating components, pages, features, fetching data, styling, routing, or working with frontend code.
debugging
Root cause analysis and debugging protocols. Use when encountering errors, test failures, unexpected behavior, stack traces, or when code behaves differently than expected.
code-review
Code review practices emphasizing technical rigor, evidence-based claims, and verification. Use when receiving code review feedback, completing tasks requiring review, or before making completion claims.
cursor-best-practices
Best practices for working with Cursor. Use when learning how to effectively use Cursor features or optimizing your workflow.
create-skill
Guide for creating effective skills following best practices. Use when creating or updating skills that extend agent capabilities.
semantic-git
Manage Git commits using conventional commit format with atomic staging. Always generate plain git commands before running them and offer to let the user run them manually.
Didn't find tool you were looking for?