vibe-ic-phase1

Missing arg? When $ARGUMENTS is empty, prompt the user first: /vibe-ic-phase1 <project-dir> (e.g. /vibe-ic-phase1 1st_benchmark_example/phase2_v0119.48-vendor). The AI must NOT guess the path; a concrete project path is required before continuing.

/phase1 — Phase 1 (Path A) entry

Main execution (program-driven):

python3 ${CLAUDE_PLUGIN_ROOT}/programs/phase1_one_shot_runner.py $ARGUMENTS

The first positional argument is the project directory (e.g. /home/<your-user>/AI_IC_design/<project>).

After the program completes, the AI must:

1. Read <project>/reports/phase1_one_shot.json — verdict + status / detail of every step
2. Complete the mandatory IC Expert track before crediting D1:
   • Read <project>/reports/audit/phase1/expert_parse_track.json.
   • If ai_subtrack.status is HANDOFF_EMITTED, use the host's Agent/subagent tool to dispatch the exact subagent_type declared in <project>/reports/audit/phase1/expert_parse_track_pack/ic_expert_agent_handoff.json. Give that agent the handoff JSON; it must read only the pack's declared design_input.txt plus expert digests and write the declared l_doc_expectations.json answer target. Do not hand-author a substitute.
   • Re-run the same phase1_one_shot_runner.py command after the agent returns. Continue this dispatch-and-rerun loop for a schema refusal; stop on an execution error. HANDOFF_EMITTED, ANSWER_SCHEMA_MISMATCH, ERROR, or CONSUMED_EMPTY is INCOMPLETE, never a Phase-1 pass.
   • Credit the track only when ai_subtrack.status == CONSUMED, ai_convergence.consumed >= 1, and both denominator.ai and denominator.total are non-zero. Preserve and report every named disagreement; findings are advisory, answer consumption is mandatory.
3. For each remaining FAIL or WAIVED step:
   • Capture the corresponding stderr / log and mark the root cause
   • If close-loop patching is needed (e.g. user did not provide input/phase1_structured.yaml), proactively guide the user via dialogue (IC Expert Agent, plain-language register) to fill in the missing info, write the dialogue result into input/phase1_structured.yaml, and re-run the runner
4. Once all PASS, print the verdict + L doc path list, and hint the next step: /phase2

Helper skills (do not auto-expand; surface only when the user asks):

• phase1 — natural-language → L1-L27 dialogue flow
• spec-review — human review after L1-L27 emit

Phase 1 not PASS means cannot enter Phase 2 — strictly depends on generated_docs/L*.json.