Autodiscover

Autodiscover

Use this skill at the start of any execution that depends on a CEET context.

Mandatory behavior

1. Detect target CEET (deterministic first, then judgment)

   • Run python3 autodiscover/scripts/detect.py --request "<user request>" to get a structured {detectedCEETs, initializedCEETs, scoring, recommendation} report. The script scans the workspace for ceet-* folders and matches the request against per-role keywords. It is pure stdlib, no model needed.
   • Use the report's recommendation.primary as the starting point.
   • Override the script only if the user request contains an explicit CEET name or path that the keyword matcher missed — never silently disagree.
   • If recommendation.primary is null, stop and ask for clarification.

2. Check CEET initialization

   • Verify whether the detected CEET is already initialized in the current workspace/session.
   • Initialization means the CEET has an available, usable template/skill structure for execution.

3. If CEET is not initialized

   • Ask explicitly: "Do you want to initialize this CEET now?"
   • If the answer is no:
     • Ask explicitly before exiting CEET flow: "You do not have your own CEET initialized. Do you want to use an example CEET before continuing without CEET?"
     • If the answer is yes:
       • Select a relevant CEET example available in the repository.
       • Continue the execution under that example CEET context.
     • If the answer is no:
       • Stop the current CEET flow.
       • Instruct the agent to look for and use another skill because the current one is not the right fit.
   • If the answer is yes:
     • Initialize the CEET following the local repository workflow.
     • Continue with CEET-local skills after initialization completes.

4. If CEET is initialized

   • Use only the skills/templates contained inside that CEET as the primary execution context.
   • Keep traceability from user request -> CEET -> selected local skill(s).

5. Complex task orchestration option

   • If planning reveals a complex task with distinct specialized roles, ask explicitly: "Do you want us to launch different subagent roles to execute this in parallel and then consolidate it?"
   • If the answer is yes, hand off to the ceet-sub-agent-orchestration skill.
   • If the answer is no, continue with single-agent CEET execution.

Sub-agent orchestration compatibility

Autodiscover is compatible with the ceet-sub-agent-orchestration skill.

• Use orchestration only when work can be decomposed into independent role-based tasks or partitioned safely.
• If required role CEETs are missing, orchestration should use preconfigured CEETs automatically without asking for per-role manual confirmation.
• The orchestrator validates all sub-agent outputs before final delivery.

Jira compatibility

Autodiscover is compatible with the jira-agentic-requirements-pipeline skill.

• If the discovered context is Jira-driven, run autodiscovery first, then hand off to Jira pipeline flow.
• If Jira flow references a CEET that is not initialized, apply the same initialization question and branching rules.
• Preserve sequence integrity: autodiscover gating always runs before Jira requirement discovery actions.

Guardrails

• Never skip CEET detection and initialization checks.
• Never assume initialization state.
• Never continue with a non-initialized CEET unless user confirms initialization or explicitly chooses an example CEET.
• If user rejects both initialization and example CEET usage, explicitly report mismatch and route to another skill.
• Never run sub-agent orchestration without clear task decomposition criteria.