AI-optimized browser automation CLI with context-efficient snapshots. Use for long autonomous sessions, self-verifying workflows, video recording, and cloud browser testing (Browserbase).
日本語の概要は準備中です。原文の説明を表示しています。
Emit and validate the provisional goal block for Autopilot Mode. Owns the 9-field format and resume detection from a pasted goal block.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Contract skill that owns the provisional goal block artifact for Autopilot Mode sessions. This skill is invoked by the orchestrator to emit, validate, and resume from the goal block. The canonical protocol lives in process/development-protocols/autopilot.md.
Invoke this skill when:
validate-autopilot-goal-block.mjs D1 validator needs to be run against an artifact.The provisional goal block — a structured text block of ≤ 4000 characters with exactly 9 named fields. The canonical field spec lives in process/development-protocols/autopilot.md §Provisional Goal Block Format. This skill does not redefine the spec — it references it.
Required fields (exact string anchors — do not rename or abbreviate):
SESSION GOAL:ENTRY PHASE:REMAINING PHASES:CLARIFICATIONS LOCKED:EXECUTE CONSENT: — must contain the literal text standing-grantedDECISION POLICY:HARD STOPS:TEST GATES:START:Step-by-step for the orchestrator:
process/development-protocols/autopilot.md §Consolidated Clarification Round).ENTRY PHASE from on-disk artifact detection (autopilot.md §Trigger-Anywhere Detection Flow).REMAINING PHASES checklist: for each phase not yet complete in canonical RIPER-5 order, add a [ ] checkbox line with the phase name and planned execution strategy.{task-folder}/{slug}_AUTOPILOT_GOAL_{dd-mm-yy}.md (header: "Emitted: [datetime]. Provisional block. V7 will emit (UPDATE) variant.").AUTOPILOT_ACTIVATED: [task] — entry phase: [phase] — goal block emitted.How the orchestrator recognizes a pasted goal block at session start:
ENTRY PHASE and REMAINING PHASES from the pasted block. Read CLARIFICATIONS LOCKED as the already-locked decisions. Read DECISION POLICY and HARD STOPS as the standing policy.[MODE: AUTOPILOT | <ENTRY PHASE>] and begin the run from START:.When VALIDATE V7 completes during an autopilot run, the orchestrator:
SESSION GOAL: with (UPDATE) .TEST GATES: TBD — populated after VALIDATE with the actual gate commands.START: to reflect post-VALIDATE state.## (UPDATE) [YYYY-MM-DD] section to the disk file (never overwrites the original).Run after writing any goal block artifact to confirm it passes D1 checks:
node .claude/skills/vc-autopilot/scripts/validate-autopilot-goal-block.mjs <artifact-path>
Exit 0 = PASS (all 9 required fields present, EXECUTE CONSENT contains standing-granted, total ≤ 4000 chars).
Exit 1 = FAIL (one or more checks failed — error printed to stdout).
WARN printed (exit 0) when TEST GATES: contains TBD (reminder that V7 UPDATE is pending).
See scripts/validate-autopilot-goal-block.mjs and fixtures/ for the D1 validator and pass/fail fixture pair.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
AI-optimized browser automation CLI with context-efficient snapshots. Use for long autonomous sessions, self-verifying workflows, video recording, and cloud browser testing (Browserbase).
日本語の概要は準備中です。原文の説明を表示しています。
Evaluate 4 execution strategies (sequential, parallel-subagents, workflow, agent-team) for a phase or fan-out task. Outputs 7-signal score table, agent count math, cost guards, and strategy recommendation.
日本語の概要は準備中です。原文の説明を表示しています。
Audit project context routing, shared-skill discoverability, and Claude/Codex wiring. Use when context docs or skill surfaces move, split, or drift.
日本語の概要は準備中です。原文の説明を表示しています。
Audit active project plan files for staleness, completion, and routing truth. Use when cleaning up plans, reconciling active work, or archiving completed artifacts.
日本語の概要は準備中です。原文の説明を表示しています。
Audit agent harness health: Claude/Codex agent parity, skill registry consistency, README.md sync, and protocol file wiring. Use when agents, skills, README.md, or development-protocol files move, split, or drift.
日本語の概要は準備中です。原文の説明を表示しています。
Loop: find gaps → fix → repeat until agents find no gaps or a metric goal is hit. Shared loop primitive for PVL, EVL, and standalone quality runs.
日本語の概要は準備中です。原文の説明を表示しています。