forgelisted

# claude-forge Orchestrator ## Step 1: Initialize or Resume **To resume a suspended pipeline**, supply the spec directory name from `.specs/`: Example: `/forge 20260401-effort-only-flow` 1. Call `mcp__forge-state__pipeline_init(arguments=$ARGUMENTS)`. 2. If `result.errors` is non-empty: surface the errors to the user and stop. 3. If `result.resume_mode` is `"auto"`: the input matched an existing spec directory. **Do not ask for confirmation** — proceed directly to Step 2. 4. For new pipelines (`resume_mode` is absent): a. If `result.fetch_needed` is non-null: 1. Fetch the external data using the method specified in `fetch_needed`: - If `fetch_needed.mcp_tool` is set: call the MCP tool with `fetch_needed.mcp_params`. - Else if `fetch_needed.command` is set: execute the command via Bash and parse the JSON output. - Else: follow `fetch_needed.instruction` as a fallback guide. 2. Map the response fields to `external_context` using `fetch_needed.response_mapping`: for each entry `(response_key → context_key)`, set `external_context[context_key] = response[response_key]`. 3. Call `mcp__forge-state__pipeline_init_with_context(workspace=result.workspace, source_id=result.source_id, source_url=result.source_url, flags=result.flags, external_context=<mapped data>)`. (`task_text` is not applicable for external issue sources — do not pass it.) b. If `result.fetch_needed` is null (plain text input): call `mc