메뉴
Track decisions (AGD) with validation and indexing. Use when making design decisions, recording important choices, discussing trade-offs, or when user mentions AGD or decision record.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Create, update, validate, diff, or export DESIGN.md files following the Google Stitch DESIGN.md spec and @google/design.md tooling. Use when the user asks to write/update DESIGN.md, create a design system document, generate design tokens, encode brand/UI guidelines, export a design system, or document UI consistency rules for AI agents. Do NOT use for ordinary UI implementation unless the user asks for DESIGN.md or design-system documentation.
Screens third-party project dependencies for known vulnerabilities and supply-chain risk before installation using a bundled check-deps.py script. Use when user asks to install, add, update, upgrade, or pin packages with npm, pnpm, yarn, bun, pip, uv, poetry, or cargo, or when deciding whether a package version is safe to install. Do NOT use for OS package managers such as apt, brew, yum, apk, or pacman.
End-to-end workflow for developing, debugging, configuring, and shipping pi-coding-agent extensions and pi packages. Use when building or maintaining extensions, skills, prompt templates, themes, settings.json setups, provider integrations, or when updating this skill after pi releases. Do NOT use for general coding questions unrelated to pi extension development.
Fork the current pi session into a new tmux pane. Use when user says /tmux-fork, 'fork to tmux', 'open in new pane', or wants to branch the conversation into a separate tmux pane.
Create and improve Claude Code Skills following official best practices. Use when creating a skill, improving skill descriptions, fixing skill activation issues, or structuring skill directories.
Convert MCP servers to Claude Code skills with progressive disclosure. Use when creating a skill from an MCP server, updating an existing MCP skill, or converting MCP documentation to skill format. Accepts MCP documentation URLs, npm package names, or direct connection info.
| name | agent-centric |
| description | Track decisions (AGD) with validation and indexing. Use when making design decisions, recording important choices, discussing trade-offs, or when user mentions AGD or decision record. |
| hooks | {"PostToolUse":[{"matcher":"Bash|Write|Edit","hooks":[{"type":"command","command":"CLAUDE_PROJECT_DIR=\"$CLAUDE_PROJECT_DIR\" \"$CLAUDE_PROJECT_DIR/.agents/scripts/validate-agds.py\""},{"type":"command","command":"CLAUDE_PROJECT_DIR=\"$CLAUDE_PROJECT_DIR\" \"$CLAUDE_PROJECT_DIR/.agents/scripts/generate-index.py\""}]}]} |
Framework for agent-centric development. Currently provides AGD (Agent-centric Governance Decision) tracking.
AGD (Agent-centric Governance Decision) is a decision record mechanism, similar to ADR or RFC. Each AGD has a unique number (e.g., AGD-001) that records important decisions, rationale, and impact.
AGD covers any important decision, not just architecture - including design patterns, conventions, tool choices, process decisions, etc.
AGD follows the RFC archival model: original files are preserved, and later decisions express how earlier ones evolve.
see-also; does not change validity of either decisionAGD exists to give the project stable references for important decisions and durable rationale that survives code evolution. It should let us trace from today's implementation back to why it was chosen, and then judge whether that decision still stands, has been updated, or has been replaced.
On each skill load, sync scripts and initialize if needed:
if [ ! -d "$CLAUDE_PROJECT_DIR/.agents" ]; then
CLAUDE_PROJECT_DIR="$CLAUDE_PROJECT_DIR" CLAUDE_SKILL_DIR="$CLAUDE_SKILL_DIR" bash "$CLAUDE_SKILL_DIR/scripts/init.sh"
else
CLAUDE_PROJECT_DIR="$CLAUDE_PROJECT_DIR" CLAUDE_SKILL_DIR="$CLAUDE_SKILL_DIR" bash "$CLAUDE_SKILL_DIR/scripts/sync-scripts.sh"
fi
If scripts were updated, briefly inform the user.
Hooks run automatically when you use Write/Edit/Bash tools on AGD files:
If validation fails, you'll see errors and should fix them (e.g., add missing tags to config.json).
AGD-{number}_{kebab-case-name}.md
Examples: AGD-001_use-postgresql.md, AGD-002_adopt-hexagonal-architecture.md
---
title: "Decision Title"
description: "Brief description"
tags: tag1, tag2
updates: AGD-001
obsoletes: AGD-002
related: AGD-003
---
## Context
Why this decision is needed.
## Decision
What was decided.
## Consequences
Impact of this decision.
See references/agd.md for complete field documentation.
IMPORTANT: Always use grep and find to search. Do NOT read files to search.
# By keyword
grep -r "keyword" "$CLAUDE_PROJECT_DIR/.agents/decisions/"
# By AGD number
find "$CLAUDE_PROJECT_DIR/.agents/decisions/" -name "AGD-001*"
# By tag
grep "#tagname" "$CLAUDE_PROJECT_DIR/.agents/INDEX-TAGS.md"
# By relationship
grep "AGD-001" "$CLAUDE_PROJECT_DIR/.agents/INDEX-AGD-RELATIONS.md"
Add tags to .agents/config.json before using them:
{
"tags": ["core", "auth", "api"]
}
See references/config.md for config details.
updated_by/obsoleted_by by syncing and pruning reverse references, skip validation of auto-generated reverse fieldsrelated relationship with RFC-style see-also semantics, extend relation index with -(r)->, clarify archival relationship model