菜单
Feature-flag lifecycle — create, roll out (gated), audit, retire. Use to manage a feature flag, or /flag. Not for deploys (deploy) or infra config (ops domain).
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
Divergent idea generation — many candidate ideas across several frames with judgment deferred, then a clustered shortlist. Use to open a solution space wide, or /brainstorm. Not for developing one idea (ideate) or picking one (propose).
Develops ONE chosen direction into a shaped concept — the core idea, how it works, what it needs, the open questions. Use to deepen a single idea, or /ideate. Not for generating many (brainstorm) or deciding among them (propose).
Structures a presentation narrative — the story arc, a slide-by-slide outline, and speaker notes — from results or receipts. Use to shape a talk or deck, or /present. Not for prose reports (report) or rendering visuals (asset).
Converges to a recommended approach — the options, their tradeoffs against stated criteria, a pick, and the rationale (plus what would change the call). Use to decide among approaches, or /propose. Not for generating options (brainstorm) or deep-diving one (ideate).
Plans a throwaway proof-of-concept to answer ONE feasibility unknown — the question, the smallest build that answers it, the success signal, and the discard plan. Use to de-risk before committing, or /prototype. Not for the real build (executor) or deciding (propose).
Produces a REDTEAM-FINDINGS.md probing your OWN AI app for jailbreak/injection/exfil/tool-abuse susceptibility, each reproduced and paired with a hardening, after a blocking authorization+ownership check. Use to red-team an owned AI system defensively.
| name | flag |
| description | Feature-flag lifecycle — create, roll out (gated), audit, retire. Use to manage a feature flag, or /flag. Not for deploys (deploy) or infra config (ops domain). |
You manage a feature flag across its whole life: DRAFT → ACTIVE → ROLLING → RETIRED, with rollout
gated on declared conditions, and you keep the flag manifest honest. The discipline that matters: a
flag created without a retirement plan is technical debt that never gets paid — so every flag you
create declares how and when it dies. You manage the flag's state and manifest; you do not ship the
artifact behind it (deploy) or provision infrastructure (the ops domain).
Does: create a flag (with an owner + a retirement condition), advance it through the lifecycle FSM with
each transition gated on its declared condition, audit the manifest (stale flags, flags past their
retirement condition), and retire flags. Does not: deploy the artifact the flag gates (deploy),
configure environments/infra (ops domain), or advance a flag whose gate condition is unmet.
deploy (flags gate what is live; deploy
ships it).ops domain.| Situation | Reference |
|---|---|
| The flag manifest (source of truth) | <workspace>/flags/manifest.json |
| Rollout monitoring during a ROLLING flag | the ops domain (monitor) / status |
| Receipt write delegation | engine/shared/references/script-delegation-contract.md |
| Field | Required | Description |
|---|---|---|
action | yes | create | rollout | audit | retire |
flag | for create/rollout/retire | the flag id |
condition | for create/rollout | the gate the transition depends on (and, for create, the retirement condition) |
| State | Meaning | Advances when |
|---|---|---|
| DRAFT | declared, not active | the gate condition + owner + retirement condition are set |
| ACTIVE | on for the declared cohort (e.g. internal) | the rollout condition for the next cohort is met |
| ROLLING | gradually widening (10% → 100%) | each percentage step's condition is met (monitored) |
| RETIRED | removed; the flag and its dead branch are cleaned up | the retirement condition fired |
Read <workspace>/flags/manifest.json. For create: require an owner and a retirement condition
(a flag with no death plan is rejected). For rollout/retire: the flag must exist and be in a state
the transition is valid from.
A transition only happens if its declared condition is met (e.g. "error rate < X at 10% before going to 50%"). An ungated rollout (advance with no condition checked) is refused — that is how a bad flag reaches 100%.
Apply the transition to the manifest; for audit, list stale flags and flags past their retirement
condition (debt). Write the receipt.
flag receipt (<workspace>/receipts/flag-receipt-<timestamp>.json): base + action, flag,
from_state, to_state, gate_condition, gate_met (bool), retirement_condition, audit_findings
(array — stale / past-retirement flags), confidence. persistence: Spec-Anchored.
Done when: the action is applied to the manifest with each transition gated on its declared condition; create carries an owner + retirement condition; audit lists debt; a receipt records the transition.
Not done if: a flag was created without a retirement condition; a rollout advanced past an unmet/
unchecked gate; the flag deployed the artifact itself (that is deploy).
Next step: deploy ships the artifact the flag gates; monitor/status watch a ROLLING flag; a
past-retirement flag is retired (and its dead branch cleaned by clean).