Files

T

YeonGyu-Kim 330e437f08 docs(agents): regenerate hierarchical AGENTS.md for 2026-05-20

Sync the AGENTS.md hierarchy to current code state:

Drift fixes in 11 existing files
- Root: 2026-05-20 commit 39aadbf9f, ~2167 TS files, 120 barrel index.ts,
  57 src/hooks dirs, 297 (179 non-test) src/shared files, 11 OpenCode hook
  handlers in plugin-interface.ts, packages list adds ast-grep-mcp + rules-core,
  first-prompt-watchdog 206 LOC, parent-wake-notifier 587 LOC
- src/AGENTS.md: file counts, plugin-interface handler count
- src/shared/AGENTS.md: title + counts 278/170 -> 297/179
- src/hooks/AGENTS.md: 57 dirs, note unwired WIP (task-reminder,
  hashline-edit-diff-enhancer)
- src/features/AGENTS.md: module map with NON-TEST counts + sub-AGENTS.md
  column, 7 modules without sub-doc
- src/features/background-agent/AGENTS.md: add 12 newer files (parent-wake-
  notifier 587 LOC, loop-detector, error-classifier, fallback-retry-handler,
  process-cleanup, subagent-spawn-limits, session-status-classifier,
  compaction-aware-message-resolver, etc.)
- src/plugin/AGENTS.md: 11 handlers, add system-transform.ts + command-
  execute-before.ts + build-team-idle-wake-hint-client.ts
- src/config/AGENTS.md: note schema/internal/permission.ts
- src/cli/AGENTS.md: 8 commands including 'version'
- src/plugin-handlers/AGENTS.md, packages/web/AGENTS.md: date bump

New AGENTS.md in 4 directories
- packages/AGENTS.md: index of 15 packages (11 platform binaries + 2 MCP
  packages + rules-core + web), role map, conventions
- docs/AGENTS.md: WHERE TO LOOK table for 19 docs across 6 subdirs
- .opencode/AGENTS.md: 5 skills + 4 slash commands + relationship to .agents/
- .agents/AGENTS.md: superset migration target (9 skills + 4 commands)

2026-05-20 17:18:44 +09:00

3.1 KiB

Raw Blame History

.opencode/ — Project-Scope Skills & Commands

Generated: 2026-05-20

OVERVIEW

Project-scope OpenCode configuration: 5 skills and 4 slash commands committed alongside the source. Picked up by src/features/opencode-skill-loader/ and the slash-command discovery pipeline.

Relationship to .agents/: .agents/ is the migration target during the oh-my-opencode → oh-my-openagent rename. It is a SUPERSET of .opencode/ (mirrors all 5 skills + adds 4 more, mirrors the 4 commands). Both directories load during the transition; consumers should prefer .agents/.

SKILLS (5)

Skill	Purpose
`work-with-pr/`	Full PR lifecycle skill: worktree → implement → atomic commits → PR → verification loop → merge.
`work-with-pr-workspace/`	Iteration workspace for `work-with-pr` — captures benchmark inputs for the iteration-* subdirs.
`github-triage/`	Read-only triage of issues + PRs. 1 item = 1 background task. Writes evidence-backed reports. Never takes GitHub actions.
`hyperplan/`	Adversarial multi-agent planning. 5 hostile category members cross-critique, lead synthesizes.
`pre-publish-review/`	Nuclear-grade 16-agent pre-publish release gate. Runs `/get-unpublished-changes`, spawns ultrabrain agents per change, oracle for synthesis.

Each skill follows the standard layout (SKILL.md + optional scripts/, references/, assets/).

COMMANDS (4 slash commands)

Command	Purpose
`/get-unpublished-changes`	Compare HEAD with the latest published npm version and list unpublished changes.
`/omomomo`	Easter-egg command — about oh-my-opencode.
`/publish`	Publish via GitHub Actions workflow. Argument: `<patch
`/remove-deadcode`	Remove unused code with LSP-verified safety + atomic commits.

CONVENTIONS

Skill YAML frontmatter is mandatory. opencode-skill-loader rejects skills without name + description.
Project-scope > user-scope. A skill at .opencode/skills/X/ overrides ~/.config/opencode/skills/X/ of the same name.
Trigger words in the skill description determine when OpenCode loads the skill. Be specific.
Commands are user-visible. Name them with a leading / (the loader normalizes filename → command).

ANTI-PATTERNS

Never duplicate a built-in skill from src/features/builtin-skills/ here without a clear reason.
Never commit background-tasks.json contents that include session IDs or token secrets.
Never write a skill that calls session.promptAsync directly — go through dispatchInternalPrompt (see root AGENTS.md "Internal message injection is dangerous").

3.1 KiB Raw Blame History