菜单
Execute exactly one lead-selected Zig actuation operation and return event-bound evidence to the coordinator. Use after $goal-actuating has prepared a capability; do not create authority, choose scope, recurse, resolve review findings, or claim goal completion.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Front door for authority-bound goal execution through the Zig actuation kernel. Bare $actuating or /goal $actuating runs implement, then $ship, then review-closeout; explicit $actuating implement stops after implementation. Explicit triage, remediation-plan, and review-closeout retain their named behavior; never infer mutation from an unqualified review request.
Translate an accepted GoalContract into normalized Zig actuation-operation topology. Use when work has repeated classes, migrations, debugging history, review campaigns, proof fanout, branch comparison, or nontrivial stopping conditions; remain advisory and never grant mutation or completion.
Fold tests, diffs, logs, benchmarks, screenshots, review results, and artifact state into a structured verdict: done, continue, regress, blocked, invalid-proof, ask-human, or refactor-kernel.
Coordinate an accepted implementation goal or explicit review workflow through Zig actuation-kernel generations. Select one bounded operation, delegate execution to $goal-grind, fold evidence, consume review-resolution/v1, acquire CAS review evidence when required, and request a Zig closure-decision/v1 without owning publication.
Compile an accepted goal, implementation spec, review campaign, migration, or hard debugging loop into a compact source-bound GoalContract. Use before multi-step execution to bind outcome, verifier, authority source, artifact scope, constraints, review policy, stop conditions, and human-owned ambiguity.
Project a source-bound GoalContract, actuation run, or review resolution into explicit inspect, edit, verify, review, ask, and block nodes. Use when decomposition changes execution; consume selected owner-boundary strategies rather than raw review findings.
| name | goal-grind |
| description | Execute exactly one lead-selected Zig actuation operation and return event-bound evidence to the coordinator. Use after $goal-actuating has prepared a capability; do not create authority, choose scope, recurse, resolve review findings, or claim goal completion. |
Consume one already-prepared capability with the smallest owner-correct action.
prepared actuation-operation/v1
-> bounded effect
-> Zig kernel observation
-> action result
-> return to $goal-actuating
selected_operation:
run_id:
step_id:
idempotency_key:
owner_boundary:
paths: []
effect: inspect | edit | verify
obligation_refs: []
capability: AKC1-...
review_resolution_ref: # review edits only
The raw capability exists only in the active executor. Never persist, quote, log, or return it. Its digest and the complete admitted operation already live in the append-only event chain.
Confirm ledger state --source actuation --run RUN_ID is prepared and its
pending step exactly matches the selected operation.
Re-read the bounded paths.
For edit, make only the owner-correct change on the admitted paths, then
consume and observe it:
ledger record --source actuation --run RUN_ID --capability "$CAPABILITY"
ledger observe --source actuation --run RUN_ID --step STEP_ID
For inspect or verify, let the kernel execute the obligation-owned
verifier:
ledger execute --source actuation --run RUN_ID --capability "$CAPABILITY"
Return changed paths, commands, transition event digests, observations, and any failure signature. Stop after this one operation.
Prefer replacement-kernel when local repair would add semantic machinery,
distribute one invariant, or preserve a dominated abstraction.
operation_result:
run_id:
step_id:
idempotency_key:
owner_boundary:
changed_paths: []
commands: []
kernel_event_refs: []
observations: []
result: passed | failed | blocked | regress
failure_signature:
public_effects: []