| source | https://github.com/juliusz-cwiakalski/agentic-delivery-os/blob/main/.opencode/README.md |
|---|
Repo-local OpenCode agents, commands, and skills.
- Agents:
.opencode/agent/*.md - Commands:
.opencode/command/*.md - Skills:
.opencode/skills/<skill-name>/SKILL.md
Note: OpenCode upstream docs use .opencode/agents/ and .opencode/commands/. This repo uses singular folders.
- Naming: kebab-case; agent/command name = filename; skill name = folder name.
- Descriptions: frontmatter
descriptionstays short (usually 3-10 words). - Commands: accept args via
$ARGUMENTSand optional$1,$2, ... - Commands: prefer
subtask: truefor non-trivial work (avoid polluting main context). - Context: keep
@pathincludes narrow; keep!shell injections small and deterministic. - Repo rules: if a tool runs repo workflows (build/test/docs), follow
AGENTS.md. - Consistency: if a new tool overlaps an existing workflow area (change lifecycle, quality gates, docs, UI), match the established patterns unless explicitly diverging.
- Prompt tuning: when updating existing tools, preserve intent and keep diffs minimal.
- Current-state prompts: describe only current required behavior, inputs, outputs, and constraints. Keep evolution/history, old names, change-specific ticket/PR references used to narrate prompt/tool evolution or rationale, and migration rationale outside prompts; runtime tracker/PR contracts and generic examples are allowed. Runtime migration, compatibility, historical-record, or prior/current-comparison tools may state only execution-essential source and target states; express accepted aliases as current input contracts where possible.
- Tool suites: when a workflow spans multiple tools, tune them together (contracts, arguments, outputs, delegation).
- Hygiene: update this file whenever you add/rename/remove a tool or materially change its intent.
- PM tracker config:
@pmreads.ai/agent/pm-instructions.md(repo-specific Jira/GitHub workflow). - PR/MR platform config:
@pr-manager,@reviewer,@review-feedback-applierread.ai/agent/pr-instructions.md(repo-specific PR/MR platform and CLI commands). - PM delegation:
@pmdelegates debugging to@fixer, commits to@committer, and command execution to@runner. - Pre-PR gate (autopilot):
@pmruns@reviewer+@doc-syncerbefore@pr-manager. - Model configuration: models are assigned in
opencode*.jsoncconfig files, NOT in agent/command definitions. Agent files describe behavior; config files define which model runs them.
decision-advisor: decisions of all types (architecture, product, business, technical, operating); decision record authoring (ADR/PDR/TDR/BDR/ODR); delegates bounded evidence gathering to@external-researcherfor selection decisionsdecision-critic: independent, read-only decision challenger; tri-state verdict (PASS / PASS_WITH_RISKS / REWORK)bootstrapper: run ADOS inception for new or legacy projectsceo: autonomous executive for ADOS delivery (Mode A) — manages the backlog, delivers tickets, verifies PM finalization, and merges approved PRs as the inner decision agent ofscripts/ceo-loop.shcoder: implement plan phases by writing code for a changecommitter: create one Conventional Commitdesigner: visual design and UI implementationdoc-syncer: reconcile system docs with changeeditor: rewrite/translate content per repo guidelinesexternal-researcher: research external sources via MCP; bounded decision-evidence pack for selection decisions (canonical-source, as-of date, data-minimization)fixer: reproduce and fix failuresimage-generator: generate AI images via text-to-image CLIimage-reviewer: analyze images, screenshots, and visual artifactsmeeting-organizer: prepare agendas and summarize meeting docsplan-writer: author change implementation planspm: orchestrate changes; manage tickets via MCP (reads.ai/agent/pm-instructions.md)pr-manager: create/update PR/MR for branch; enriches description with ticket context via MCPreadiness-reviewer: adversarially critique change artifacts (spec/test-plan/plan) against the source ticket before delivery (Definition of Ready gate)review-feedback-applier: classify and apply accepted review feedback from PR/MRreviewer: review changes against spec, plan, code quality heuristics, and repo rules (local + remote modes)runner: run commands and capture logsspec-writer: author change specificationstest-plan-writer: author change test planstoolsmith: create and tune OpenCode tooling
/apply-review-feedback: classify and apply accepted PR/MR review feedback locally (via@review-feedback-applier)/bootstrap: run ADOS project inception/check: run quality gates (no fixes)/check-fix: run quality gates and fix failures/check-readiness: run the Definition of Ready gate for a change (via@readiness-reviewer)/commit: create one Conventional Commit/design: generate/update visual design assets/plan-change: plan a change (prep context)/plan-decision: interactive decision session (any type: architecture, product, business, technical, operating)/pr: create/update PR/MR and sync title/description (tmp/pr/<branch>/description.md, via@pr-manager); fetches ticket context from Jira/GitHub whenworkItemRefis detected/review: review a change vs spec/plan/review-deep: deeper review vs spec/plan/review-decision: independent decision challenge (delegates to@decision-critic)/review-remote: review open PR/MR diff and optionally publish findings (via@reviewerremote mode)/run-plan: execute an implementation plan/sync-docs: reconcile system specs from a change/write-decision: write a decision record (ADR/PDR/TDR/BDR/ODR) from planning context/write-plan: generate an implementation plan/write-spec: generate a change spec/write-test-plan: generate a change test plan