Hep Storm

Run the Stormbreaker Loop over the fabric

Execute the goal to completion under this protocol. Do not stop to ask for confirmation — this is a force-robust run. Only halt when the goal is verified, or you are genuinely blocked by auth, payment, policy, a missing secret/tool, or a required user approval.

1. scope-lock — Restate the goal as one sentence and lock to it. Check the route decision's failure-memory. If (and only if) the goal is too ambiguous to decompose safely, ask ONE batch of 3–5 questions (what NOT to do / smallest version / done signal / dependencies), then proceed. If it is already specific, ask nothing.
2. issue contract — Write the acceptance criteria: the concrete, checkable done-signal for each packet and for the whole goal. These become the loop's stop criteria.
3. plan-lock — Adopt the fabric's parallel_groups and depends_on ordering as the plan. Open a visible goal ledger: packet, owner, verification gate, status, resume point.
4. act — Execute the next unblocked group. Run independent packets in the group concurrently where the runtime supports it. When a packet's card names an Agentlas specialist, borrow and run it attached to this project via "$RUNNER" hep-call "<card>" "<goal>" --project . rather than role-playing it. Write artifacts to each packet's write_scope.
5. verify — A packet passes only with separately attributable, validated verification. A loop.goal_command exiting 0 may end the goal loop, but it cannot verify the executor's own output or self-attest success. Require an independent verifier and hash-validated artifact evidence. "It ran" is never success.
6. bounded repair/retry — On a concrete validation failure, repair and re-run that packet — bounded. Honor the goal-loop budget: tolerate transient failures with backoff, stop a packet as stalled after consecutive no-progress iterations, and never exceed its iteration ceiling. A resumed run continues from the journal, it does not restart.
7. final-gate — Report success only if the fabric's final gate clears (can_report_success). Run the evidence/security gate on produced artifacts.

Keep visible progress concise: what was attempted, what was verified, and exactly where to resume if blocked. Never expose hidden reasoning — show progress, evidence, decisions, and final status only.

Loop invariants (why this beats a one-shot loop)

• Don't break (안 끊기게): a transient packet failure is journaled and retried with backoff, not fatal. Only a genuine streak of hard failures stops the run.
• Don't run away: a hard iteration ceiling plus stall detection — measured no-progress stops the loop as stalled instead of spinning.
• Keep the goal until done (될 때까지): the loop reports reached_goal only when the verifier proves it. No bare "it ran" ever counts.
• Survive a hard stop: every packet is a journal step, so a killed run resumes its numbering from the journal instead of colliding or restarting from zero.

Hard rules

• No fake pass. If the engine is unavailable, an account/tool/connector/browser session is missing, or a gate did not run, report the run as blocked or unverified with the exact next step — never as complete. A scheduled or materialized run is not proof that an external action succeeded.
• The router only chooses agents and fetches BYOM bundles. Actual tool execution follows this runtime's own safety and permission model.
• Report the receipt_id, pipeline_id, and journal path in your final message so the run is auditable and resumable.

Examples

/hep-storm ship a working waitlist landing page with a verified signup flow
/hep-storm 이 리포 결제 버그를 재현 PoC까지 만들어서 고치고 회귀 테스트로 검증해줘
@Hephaestus storm turn this research question into a cited report with evidence

Update fallback: 자동 업데이트가 안 되면 hephaestus update를 한 번 실행하세요. 업데이트하지 않아도 현재 버전 명령은 그대로 동작합니다.