Reusable agent skills and workflows for everyday development.
Before installing an agent, review the prerequisites for required CLI tools, terminal setup, and provider access.
- Claude Code - Anthropic’s terminal-native coding agent.
- revdiff plugin - Interactive diff review and annotation workflow.
- Codex plugin for Claude Code - Adds OpenAI Codex integration to Claude Code.
- Pi Coding Agent - Extensible terminal coding agent with packages, tools, skills, and TUI support.
- revdiff package - Interactive code review and diff navigation.
- pi-subagents package - Delegates work to focused subagents and chains.
- pi-web-access package - Adds web search, content fetching, and library research tools.
- pi-intercom package - Lets multiple Pi sessions communicate locally.
- pi-review-loop package - Automates repeated code-review passes until issues are resolved.
To use coding-harness/skills in your projects, copy or symlink the skills folder to:
- For Claude Code:
project-folder/.claude/skillsor~/.claude/skills
- For Pi Coding Agent, OpenCode, Codex, Copilot:
project-folder/.agents/skillsor~/.agents/skills
Ask your coding agent in natural language or use the skill command syntax supported by your host. The examples below use portable prompts; direct skill invocations use /<skill-name> explicitly.
autopilot is manual-only. Invoke it explicitly when the objective has checkable completion predicates and the user wants restart-resumable execution through a supported coding harness:
/autopilot I am going to sleep. Work overnight on migrating this multi-module Java backend
from Spring Boot 3.x to Spring Boot 4.0.0. Done means the Maven verify build,
integration tests, and application-context smoke tests pass; no Spring Boot 3.x
artifacts remain; and the migration guide is updated. Open a PR or MR, but do not merge or deploy.
Autopilot requires Node.js 24+ and Git. Pi can load the packaged Autopilot extension and an active pi-subagents owner for process-local structured execution; an unavailable owner selects the distinct direct Pi CLI fallback before admission. Autopilot never installs or enables these components. Remote delivery also requires gh or glab and explicit per-run grants.
Use /autopilot status to check overnight progress, /autopilot resume to continue an interrupted nonterminal run, or /autopilot stop to end a run while preserving its work. /autopilot address review comments snapshots feedback from the exact open PR/MR, creates a sealed amendment successor, and resolves provider-resolvable threads only after the fix passes. After provider-confirmed merge, /autopilot wrap up performs guarded remote-branch, sibling-worktree, local-branch, and amendment-chain cleanup; use /autopilot wrap up with handoff to preserve project-local summaries.
Activate the brainstorm skill:
let's brainstorm this user story: """User story description"""
If brainstorming produced an implementation plan, use the project-appropriate coding skill, such as java-coder or typescript-coder:
implement the plan YYYY-MM-DD-user-story-implementation-plan.md
using the applicable project coding skill and provide human-verifiable evidence
The architect skill is manual-only. Use it after behavior and important constraints are understood to produce a caller-first, implementation-ready structural design.
/architect Define the types, signatures, and module boundaries for this approved design.
It returns a design by default. Implementation requires an explicit request or later approval.
The arena skill is manual-only for direct use. Invoke it when several independent attempts at the same bounded artifact would improve the decision.
/arena Produce three independent designs for this parser contract and synthesize the strongest result.
It returns one coherent synthesis and the applicable verification result rather than a collection of candidate fragments.
Activate the code-review skill:
code review uncommitted changes
The blast-radius skill is manual-only. Invoke it with a bounded local comparison:
/blast-radius Analyze HEAD~1..HEAD.
It traces credible effects beyond the obvious diff and seeks executable proof for decision-critical safety assumptions.
The reflect skill is manual-only. Invoke it after a substantial session:
/reflect Review this session and propose evidence-backed improvements.
It proposes evidence-backed improvements to existing skills and waits for individual approval before editing.
The teach skill is manual-only. Invoke it with the learning request:
/teach Explain how request retries work in this service.
It explains repository-backed code progressively without modifying the project.
Activate the write-tests and test-coverage skills:
write tests for new and modified code
add test and branch coverage for UserController
Activate the debug-error and/or clarify skills:
debug this error: """stack trace or errors in logs"""
clarify why the response status is 500 instead of 201
Activate the refactor-code and java-coder skills:
refactor code in the user login flow, following guidelines from the java-coder skill
Use the technical-writing skill for substantial technical documents and design artifacts:
review this deployment runbook for incorrect assumptions, missing prerequisites, and unsafe recovery steps
Use unslop alone for small style-only edits.
Choose the diagram skill based on the artifact you need:
- Use
pretty-mermaidfor Mermaid-as-code, themed SVG rendering, ASCII output, or batch rendering. - Use
diagram-designfor editorial standalone HTML/SVG diagrams, branded styling, or redraws of Mermaid and draw.io sources.
Use the pretty-mermaid skill to create an ASCII terminal-friendly sequence diagram of the endpoint /api/v1/users.
Use the diagram-design skill to create an editorial architecture diagram of this service as standalone HTML.