Specify a webhook-driven task for OpenClaw
Describe event inputs, validation, deduplication, and intended actions before integration work.
AOSP investigation-driven planning with parallel code search and optional Codex integration
--- name: aosp-plan-bf1b1822bcfc description: AOSP investigation-driven planning with parallel code search and optional Codex integration argument-hint: <AOSP investigation query> level: 4 --- # AOSP Plan Skill Investigation-first AOSP planning. Decomposes queries into facets, spawns parallel `aosp-investigator` subagents, optionally integrates Codex lanes, synthesizes all findings, and produces an evidence-backed plan saved to `.omc/plans/`. ## Usage ``` /oh-my-claudecode:aosp-plan "query about AOSP code" /oh-my-claudecode:aosp-plan --agents 5 "query" /oh-my-claudecode:aosp-plan --codex "query" /oh-my-claudecode:aosp-plan --agents 4 --codex "query" ``` ## Flags - `--agents N`: Number of parallel investigator subagents (default: 3, max: 5) - `--codex`: Enable additional Codex investigation lanes via `omc ask codex` ## Protocol ### Step 1: MCP Health Check Call `aosp_code_search` with `tool: "list_tools"` once at startup to verify the MCP server is reachable and discover available remote tool names. If the call fails, abort immediately with: ``` AOSP MCP server unreachable. Check AOSP_MCP_URL and AOSP_MCP_KEY environment variables. ``` Do NOT proceed to spawn agents if this check fails. ### Step 2: Facet Decomposition Given the user query, decompose into 2–N independent investigation facets. Each facet targets a different aspect of the AOSP codebase. Show the decomposition to the user: ```markdown ## AOSP Investigation Decomposition **Query:** <original query> ### Facet 1: <facet-name> - **Search focus:** What to search for in AOSP - **Expected areas:** Framework, HAL, kernel, etc. ### Facet 2: <facet-name> - **Search focus:** ... - **Expected areas:** ... ``` ### Step 3: Spawn Investigators Fire N `aosp-investigator` subagents in parallel (one per facet). N comes from `--agents` (default 3, max 5): ``` Agent( subagent_type="oh-my-claudecode:aosp-investigator", model="sonnet", prompt="Investigate AOSP facet: <facet description>. Use aosp_code_search tool. Report structured findings with file paths, code snippets, and architectural observations." ) ``` Cap at 5 agents regardless of `--agents` value. ### Step 4: Codex Lanes (--codex only) If `--codex` flag is set, spawn additional investigation lanes for each facet via Codex. These run in parallel with the Claude investigators and are additive (not counted against the agent cap): ```bash omc ask codex "Analyze AOSP code for: <facet description>. Focus on: <specific aspects>. Report file paths, code patterns, and architectural notes." ``` Codex artifacts land in `.omc/artifacts/ask/codex-*.md`. ### Step 5: Codex Result Normalization (--codex only) Read all Codex artifacts from `.omc/artifacts/ask/codex-*.md`. Normalize each into the same structured format as Claude investigator results: ```markdown ### Codex Findings: <facet> **Source:** Codex analysis **Findings:** - <finding with file path and context> - <finding with file path and context> **Architectural Notes:** - <observation about code structure, design patterns, or AOSP conventions> ``` ### Step 6: Synthesis Merge all investigation results (Claude investigators + Codex findings if applicable): - Deduplicate overlapping findings across sources - Resolve conflicts between sources (prefer AOSP source-based evidence) - Rank findings by relevance and evidence strength - Note gaps where investigation was inconclusive or returned no results ### Step 7: Plan Generation Generate a structured plan based on investigation evidence: ```markdown # AOSP Plan: <query> ## Investigation Summary [Key findings from all investigators, grouped by facet] ## Evidence-Based Plan ### Step 1: <action> - **Evidence:** [which investigation findings support this step] - **AOSP files:** [relevant AOSP source files] - **Acceptance criteria:** [how to verify this step is complete] ### Step 2: <action> - **Evidence:** ... - **AOSP files:** ... - **Acceptance criteria:** ... ## Open Questions [Items where investigation was inconclusive or returned no results — flag for further investigation] ## Sources [All AOSP file paths and references cited across all investigator results] ``` ### Step 8: Save Derive a slug from the query (lowercase, spaces→hyphens, strip special chars). Save to: ``` .omc/plans/aosp-<slug>.md ``` Confirm the save path to the user after writing. ## Configuration - Maximum 5 parallel `aosp-investigator` agents (matches `external-context` precedent) - Codex lanes are additive (not counted against the agent limit) - Keyword trigger: `"aosp plan"` or `"aosp_plan"`--- name: aosp-plan-bf1b1822bcfc description: AOSP investigation-driven planning with parallel code search and optional Codex integration argument-hint: <AOSP investigation query> level: 4 --- # AOSP Plan Skill Investigation-first AOSP planning. Decomposes queries into facets, spawns parallel `aosp-investigator` subagents, optionally integrates Codex lanes, synthesizes all findings, and produces an evidence-backed plan saved to `.omc/plans/`. ## Usage ``` /oh-my-claudecode:aosp-plan "query about AOSP code" /oh-my-claudecode:aosp-plan --agents 5 "query" /oh-my-claudecode:aosp-plan --codex "query" /oh-my-claudecode:aosp-plan --agents 4 --codex "query" ``` ## Flags - `--agents N`: Number of parallel investigator subagents (default: 3, max: 5) - `--codex`: Enable additional Codex investigation lanes via `omc ask codex` ## Protocol ### Step 1: MCP Health Check Call `aosp_code_search` with `tool: "list_tools"` once at startup to verify the MCP server is reachable and discover available remote tool names. If the call fails, abort immediately with: ``` AOSP MCP server unreachable. Check AOSP_MCP_URL and AOSP_MCP_KEY environment variables. ``` Do NOT proceed to spawn agents if this check fails. ### Step 2: Facet Decomposition Given the user query, decompose into 2–N independent investigation facets. Each facet targets a different aspect of the AOSP codebase. Show the decomposition to the user: ```markdown ## AOSP Investigation Decomposition **Query:** <original query> ### Facet 1: <facet-name> - **Search focus:** What to search for in AOSP - **Expected areas:** Framework, HAL, kernel, etc. ### Facet 2: <facet-name> - **Search focus:** ... - **Expected areas:** ... ``` ### Step 3: Spawn Investigators Fire N `aosp-investigator` subagents in parallel (one per facet). N comes from `--agents` (default 3, max 5): ``` Agent( subagent_type="oh-my-claudecode:aosp-investigator", model="sonnet", prompt="Investigate AOSP facet: <facet description>. Use aosp_code_search tool. Report structured findings with file paths, code snippets, and architectural observations." ) ``` Cap at 5 agents regardless of `--agents` value. ### Step 4: Codex Lanes (--codex only) If `--codex` flag is set, spawn additional investigation lanes for each facet via Codex. These run in parallel with the Claude investigators and are additive (not counted against the agent cap): ```bash omc ask codex "Analyze AOSP code for: <facet description>. Focus on: <specific aspects>. Report file paths, code patterns, and architectural notes." ``` Codex artifacts land in `.omc/artifacts/ask/codex-*.md`. ### Step 5: Codex Result Normalization (--codex only) Read all Codex artifacts from `.omc/artifacts/ask/codex-*.md`. Normalize each into the same structured format as Claude investigator results: ```markdown ### Codex Findings: <facet> **Source:** Codex analysis **Findings:** - <finding with file path and context> - <finding with file path and context> **Architectural Notes:** - <observation about code structure, design patterns, or AOSP conventions> ``` ### Step 6: Synthesis Merge all investigation results (Claude investigators + Codex findings if applicable): - Deduplicate overlapping findings across sources - Resolve conflicts between sources (prefer AOSP source-based evidence) - Rank findings by relevance and evidence strength - Note gaps where investigation was inconclusive or returned no results ### Step 7: Plan Generation Generate a structured plan based on investigation evidence: ```markdown # AOSP Plan: <query> ## Investigation Summary [Key findings from all investigators, grouped by facet] ## Evidence-Based Plan ### Step 1: <action> - **Evidence:** [which investigation findings support this step] - **AOSP files:** [relevant AOSP source files] - **Acceptance criteria:** [how to verify this step is complete] ### Step 2: <action> - **Evidence:** ... - **AOSP files:** ... - **Acceptance criteria:** ... ## Open Questions [Items where investigation was inconclusive or returned no results — flag for further investigation] ## Sources [All AOSP file paths and references cited across all investigator results] ``` ### Step 8: Save Derive a slug from the query (lowercase, spaces→hyphens, strip special chars). Save to: ``` .omc/plans/aosp-<slug>.md ``` Confirm the save path to the user after writing. ## Configuration - Maximum 5 parallel `aosp-investigator` agents (matches `external-context` precedent) - Codex lanes are additive (not counted against the agent limit) - Keyword trigger: `"aosp plan"` or `"aosp_plan"`
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.