From crucible
Resume a halted /crucible:forge or /crucible:autopilot run by inspecting the evidence tree and continuing from the first missing phase artifact. Evidence-tree-as-state — no separate state file required.
npx claudepluginhub krzemienski/crucible --plugin crucible# /crucible:resume A forge or autopilot run can halt mid-pipeline (network blip, manual interrupt, context overflow). Resume picks up where the last run left off by inspecting the evidence tree itself — the evidence IS the state. ## Pipeline ### Step 1 — Identify the most recent run Find the most recent run-id directory across: - `evidence/oracle-plan-reviews/` - `evidence/validation-artifacts/` - `evidence/robust-trials/` Use the latest mtime as the active run. Print: ### Step 2 — Probe each forge phase artifact Walk the 10 forge phases and check for the expected artifact: | Phase |...
/resumeResumes a previous Claude agent by ID, loading its transcript to continue the interrupted task where it left off. Pass agent ID; optionally add follow-up prompt.
/resumeLists active task orchestrations with progress and resumes specific ones via --date --project or --latest. Shows tasks in progress, blocked/next tasks, git activity, and quick actions.
/resumeLists active task orchestrations with progress and resumes specific ones via --date --project or --latest. Shows tasks in progress, blocked/next tasks, git activity, and quick actions.
/resumeResumes a babysitter workflow run by run ID or discovers/suggests incomplete runs based on status, inputs, and process.
/resumeResumes interrupted Lisa specification interviews: scans .claude/lisa-*.md state files, lists for user selection, continues questions via AskUserQuestion, updates draft spec, finalizes MD/JSON docs.
/resumeDetects and resumes interrupted Plan Cascade tasks by auto-detecting mega-plan, hybrid-worktree, or hybrid-auto context and routing appropriately.
A forge or autopilot run can halt mid-pipeline (network blip, manual interrupt, context overflow). Resume picks up where the last run left off by inspecting the evidence tree itself — the evidence IS the state.
Find the most recent run-id directory across:
evidence/oracle-plan-reviews/evidence/validation-artifacts/evidence/robust-trials/Use the latest mtime as the active run. Print:
Active run: <run-id> (started <iso timestamp>)
Walk the 10 forge phases and check for the expected artifact:
| Phase | Artifact path | Expected content |
|---|---|---|
| 1 | evidence/codebase-analysis/<run>/SUMMARY.md | non-empty |
| 2 | evidence/documentation-research/SUMMARY.md | non-empty |
| 3 | evidence/oracle-plan-reviews/<run>/plan.md | non-empty + has MSC- lines |
| 4 | evidence/oracle-plan-reviews/<run>/oracle-1-verdict.md | contains APPROVE or BLOCK |
| 5 | evidence/session-receipts/ (latest) | recent receipts for this run |
| 6 | evidence/validation-artifacts/<run>.md | non-empty |
| 7 | evidence/INDEX.md | mtime ≥ Phase 6 artifact |
| 8 | evidence/reviewer-consensus/decision.md | contains UNANIMOUS PASS or FAIL |
| 9 | evidence/final-oracle-evidence-audit/decision.md | contains APPROVED or BLOCK |
| 10 | evidence/completion-gate/report.json | parses; has overall key |
The first phase whose expected artifact is missing or empty is the resume point.
Invoke the forge pipeline starting at the resume phase. Skip earlier phases (their artifacts are already valid). Continue through Phase 10.
If no phase is missing (all 10 artifacts present), check the gate report:
overall=COMPLETE: report "already complete; nothing to resume." Stop.overall=REFUSED: invoke /crucible:remediate, then re-run forge from Phase 5.CRUCIBLE RESUME
===============
Active run: <run-id>
Phases complete: 1, 2, 3, 4
Resuming at: Phase 5 (Execute)
Reason: evidence/session-receipts/ has no entries newer than plan.md
After completion, output mirrors /crucible:forge (COMPLETE or REFUSED).