Adonis0123/adonis-skills / skills/goal-gate
goal-gate
VerifiedGate whether a coding-agent task benefits from a durable, verifiable contract, then draft, validate, start, continue, or close it for Codex, Grok, Claude Code, Cursor CLI, or an unknown host. Use for explicit goal or /goal requests and autonomous multi-checkpoint coding work with one checkable end state. Native Goal creation or mutation requires an explicit user or system request. Do not use for quick one-shot work, unrelated backlogs, OKRs, reminders, or token-budget-only changes.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 14.5 KB |
| agents/openai.yaml | config | 386 B |
| evals/evals.json | config | 46.5 KB |
| evals/trigger-eval.json | config | 3.7 KB |
| references/copy-ready-goals.md | markdown | 10.9 KB |
| references/examples.md | markdown | 26.7 KB |
| references/runtime-actions.md | markdown | 10.8 KB |
| scripts/lint-goal-prompt.pyexecutable | script | 6.2 KB |
| scripts/test-lint-goal-prompt.pyexecutable | script | 10.8 KB |
references/examples.md
View on GitHub# Goal Gate Examples
High fit authorizes the work under a transcript contract; native Goal mutation still needs explicit Goal authorization or an already-authorized continuation. Medium fit stays on `suggest`; a tripped safety gate drops to `suggest` or `defer`.
## Grok tooling: implicit work adopts without native Goal
User: "I'm in Grok Build. The session has `update_goal`. No goal is active. Work through PLAN.md and don't stop until `pnpm test` passes and `pnpm build` succeeds."
High fit, gate clear, but the user did not request native Goal state. Adopt the contract and continue; do not emit `/goal`, do not call `update_goal` while inactive, and do not claim product Goal is Active.
```text
Goal Gate
- Decision: suggest
- Runtime: grok-tooling
- Goal fit: high
- Objective: Implement every task in PLAN.md until the named validation commands pass.
- Done condition: All PLAN.md tasks are implemented, `pnpm test` passes, and `pnpm build` succeeds.
- Verification: Surface changed files, the test/build commands, and their exit status.
- Constraints: Stay within PLAN.md scope; do not push or commit unless asked.
- Execution strategy: Before implementation, assess plan-task independence, shared context, write overlap, and separate verification; use subagents only when beneficial and keep final integration with the main agent.
- Checkpoints: Report in transcript only while the product Goal is inactive.
- Stop or ask when: PLAN.md is ambiguous, validation needs unavailable credentials, or a destructive action is required.
- Prompt: none
- Next: adopt goal and continue
```
## Grok tooling: explicit Goal request waits for user /goal
User: "I'm in Grok Build. The session has `update_goal`. No goal is active. Set a durable goal and work through PLAN.md until `pnpm test` passes and `pnpm build` succeeds."
High fit, gate clear, and the user explicitly requested native Goal state. Grok has no `create_goal` — durable mode is inactive until the user pastes `/goal`. Emit the contract + copy-ready slash command, set `Next: wait for user /goal`, and **stop**. Do not start PLAN.md work and do not call `update_goal` yet.
```text
Goal Gate
- Decision: set-now
- Runtime: grok-tooling
- Goal fit: high
- Objective: Implement every task in PLAN.md until the named validation commands pass.
- Done condition: All PLAN.md tasks are implemented, `pnpm test` passes, and `pnpm build` succeeds.
- Verification: Surface changed files, the test/build commands, and their exit status.
- Constraints: Stay within PLAN.md scope; do not push or commit unless asked.
- Execution strategy: Before implementation, assess plan-task independence, shared context, write overlap, and separate verification; use subagents only when beneficial and keep final integration with the main agent.
- Checkpoints: After goal is Active, report after each PLAN.md milestone via `update_goal` message.
- Stop or ask when: PLAN.md is ambiguous, validation needs unavailable credentials, or a destructive action is required.
- Prompt: see Recommended /goal below
- Next: wait for user /goal
```
```text
/goal Implement every task in PLAN.md without stopping until pnpm test passes and pnpm build succeeds. Execution strategy: assess dependencies, shared context, write overlap, and independent verification before using subagents; keep final integration with the main agent. Surface each checkpoint with commands and exit codes. Pause for credentials, production data, or destructive actions.
```
One-line truth for the user: `/goal-gate` drafted this contract; paste the `/goal` line above (or say "soft only / no native goal") to continue. Do not call `create_goal` or `get_goal`. After the user activates, work under Active goal and call `update_goal` with checkpoint `message` values; call `completed: true` only after verification evidence proves the done condition; call `blocked_reason` only when genuinely stuck.
## Grok tooling: set-now after user activates /goal
User follows the previous turn by pasting the recommended `/goal …` (or `/goal status` shows Active).
Now work the contract. Checkpoints use `update_goal` messages; completion uses `completed: true` only with surfaced evidence. If `update_goal` returns `Goal is not Active`, stop and re-request `/goal` activation — do not retry complete.
## Grok tooling: soft-adopt exception (user declines native goal)
User: "I'm in Grok. Just do PLAN.md until tests pass — no /goal, soft only."
```text
Goal Gate
- Decision: set-now
- Runtime: grok-tooling
- Goal fit: high
- Objective: Implement every task in PLAN.md until the named validation commands pass.
- Done condition: All PLAN.md tasks are implemented, `pnpm test` passes, and `pnpm build` succeeds.
- Verification: Surface changed files, the test/build commands, and their exit status.
- Constraints: Stay within PLAN.md scope; soft-adopt only — no durable Grok goal.
- Execution strategy: Assess plan-task independence before subagents; main agent keeps final integration.
- Checkpoints: Report in transcript only (no update_goal while inactive).
- Stop or ask when: PLAN.md is ambiguous or a destructive action is required.
- Prompt: none (user declined durable /goal)
- Next: adopt goal and continue
```
Never call `update_goal` while inactive, even on soft-adopt.
## Grok slash: copy-ready Chinese prompt
User: "帮我写一个 Grok /goal,我想做一个本地待办 App,先给能直接复制的版本。"
```text
Goal Gate
- Decision: suggest
- Runtime: grok-slash
- Goal fit: high
- Objective: Build a first-version local personal todo app MVP.
- Done condition: The local app runs and the core add/list/complete todo workflow is proven with runtime evidence.
- Verification: Surface project commands, local run output, and browser evidence for the workflow.
- Constraints: No accounts, paid services, production deploy, or cloud sync.
- Execution strategy: Assess decomposability before implementation; keep final integration with the main agent.
- Checkpoints: Report after inspection, first runnable workflow, and final verification.
- Stop or ask when: Credentials, deployment, or product-scope decisions are required.
- Prompt: see Recommended /goal below
- Next: provide prompt
```
```text
推荐执行版(中文,可直接复制)
/goal 创建第一版本地待办 App MVP;先读取项目已有命令和约束,若无现有项目则创建最小本地 Web 应用,实现添加、列表、完成/取消完成待办的核心流程。
验证:运行项目最小相关检查,启动本地应用,在浏览器走通新增、列表、完成待办,并用命令输出或截图作为证据。
约束:不接入账号、云同步、付费服务或生产部署。
边界:只写入新项目目录,或只修改与待办核心流程直接相关的文件。
执行编排:开始前评估依赖、共享上下文、写入冲突与独立验证能力再决定是否用 subagent;主 agent 负责最终集成验证。
迭代策略:先做可运行核心流程,最多 3 轮聚焦改进。
完成条件:本地核心待办流程可运行且验证证据已展示。
暂停条件:需要凭证、付费、生产部署或产品范围决策时暂停。
默认选择理由:本地 MVP 最快验证核心流程,并避开账号与云端风险。
```
Note: `/goal` in Grok is available when the goal feature is enabled and `update_goal` is in the toolset. Management: `/goal status` | `pause` | `resume` | `clear`.
## Grok tooling: active goal already present
User: "I'm in Grok. A goal is already active for the search-indexing refactor. Set a new goal to finish the pagination rewrite."
```text
Goal Gate
- Decision: defer
- Runtime: grok-tooling
- Goal fit: high
- Objective: Finish the pagination rewrite with validation evidence.
- Done condition: Pagination rewrite complete and agreed checks pass.
- Verification: Surface active-goal status, proposed objective, and validation output once approved.
- Constraints: Do not clear, complete, or replace the active goal without user approval.
- Execution strategy: n/a until the active-goal decision is resolved.
- Checkpoints: n/a
- Stop or ask when: A goal is already active.
- Prompt: none
- Next: ask approval
```
Ask whether to continue the active goal, mark it complete via `update_goal` only if truly done, `/goal pause|resume|clear`, or replace after clear. Do not call `update_goal(completed=true)` on the wrong objective.
## Codex slash: vague Chinese app prompt
User: "帮我写一个 Codex /goal,我想做一个本地记账 App,先给能直接复制的版本。"
The user wants a prompt, not immediate execution. Use conservative defaults and make the first `/goal` copy-ready.
```text
Goal Gate
- Decision: suggest
- Runtime: codex-slash
- Goal fit: high
- Objective: Build a first-version local personal budgeting app MVP.
- Done condition: The local app runs and the core add/view/edit transaction workflow is proven with runtime evidence.
- Verification: Surface discovered project commands, local run output, screenshots or browser evidence for the workflow, and any skipped checks.
- Constraints: No account system, paid services, production deployment, bank integration, or real financial advice.
- Execution strategy: Assess decomposability before implementation; delegate only bounded, independently verifiable work and keep final integration with the main agent.
- Checkpoints: Report after project inspection, first runnable workflow, and final verification.
- Stop or ask when: Credentials, bank APIs, deployment, regulated financial advice, or unclear data ownership is required.
- Prompt: see Recommended /goal below
- Next: provide prompt
```
```text
推荐执行版(中文,可直接复制)
/goal 创建第一版本地个人记账 App MVP,先读取项目已有命令和约束;如果没有现有项目,就创建最小本地 Web 应用,实现添加、查看、编辑和删除一笔收支记录的核心流程。
验证:运行项目提供的最小相关检查,启动本地应用,在浏览器中完整走通新增、查看、编辑、删除收支记录流程,并用命令输出、日志或截图作为证据。
约束:不接入真实银行、支付、账号系统、云同步、生产部署或金融建议;不保存真实敏感财务数据。
边界:只写入新项目目录,或只修改现有项目中与记账核心流程直接相关的文件。
执行编排:开始实现前结合任务复杂度、依赖、共享上下文、写入冲突和独立验证能力判断是否使用 subagent;主 agent 负责审查结果和最终集成验证,不适合或不支持时退化为单 agent。
迭代策略:先做可运行核心流程,再根据检查结果和浏览器证据做最多 3 轮聚焦改进;同一错误连续失败 2 次后换日志、控制台或文档证据来源。
完成条件:本地核心记账流程可运行,验证证据已展示,检查通过或缺失配置已明确说明。
暂停条件:需要真实银行接口、账号凭证、付费服务、生产部署、法律/金融判断、敏感真实数据或产品范围决策时暂停。
默认选择理由:先做本地 MVP 能最快验证核心记账流程,同时避开账号、银行接口和生产数据风险。
```
## Codex tooling: implicit work adopts without native Goal
User: "I'm in Codex with goal tooling. No goal is active. Work through PLAN.md and don't stop until `pnpm test` passes and `pnpm build` succeeds."
High fit and the safety gate is clear, but the user did not request native Goal state. Adopt the contract and continue; do not call `create_goal` or claim product Goal is Active.
```text
Goal Gate
- Decision: suggest
- Runtime: codex-tooling
- Goal fit: high
- Objective: Implement every task in PLAN.md until the named validation commands pass.
- Done condition: All PLAN.md tasks are implemented, `pnpm test` passes, and `pnpm build` succeeds.
- Verification: Surface changed files, the test/build commands, and their exit status.
- Constraints: Stay within PLAN.md scope; do not push or commit unless asked.
- Execution strategy: Before implementation, assess plan-task independence, shared context, write overlap, and separate verification; use installed subagent orchestration only when it helps, with the main agent retaining final integration responsibility.
- Checkpoints: Report after each PLAN.md milestone in the transcript.
- Stop or ask when: PLAN.md is ambiguous, validation needs unavailable credentials, or a destructive action is required.
- Prompt: none
- Next: adopt goal and continue
```
## Codex tooling: explicit set-now
User: "Set a goal to finish PLAN.md and keep going until validation passes."
Here the user explicitly authorizes native Goal creation, so the runtime action differs from the implicit-work example.
```text
Goal Gate
- Decision: set-now
- Runtime: codex-tooling
- Goal fit: high
- Objective: Implement PLAN.md until the required validation commands pass.
- Done condition: Every PLAN.md acceptance criterion is implemented and the named validation commands exit 0.
- Verification: Surface changed files, validation commands, exit status, and any skipped checks.
- Constraints: Follow PLAN.md scope; do not push or commit unless explicitly asked.
- Execution strategy: Assess single-agent versus delegated execution from plan dependencies and write boundaries; parallelize only independent non-conflicting tasks and keep final verification with the main agent.
- Checkpoints: Report after each plan milestone.
- Stop or ask when: PLAN.md is ambiguous, validation requires unavailable credentials, or a destructive action is needed.
- Prompt: none
- Next: create goal
```
## Codex tooling: safety gate holds (auth)
User: "Migrate our auth module from server sessions to JWT across the Node backend. Set this up to run to completion."
High fit and tooling is available, but auth work trips the safety gate, so do not auto-create the goal — ask first.
```text
Goal Gate
- Decision: suggest
- Runtime: codex-tooling
- Goal fit: high
- Objective: Migrate auth from server sessions to JWT while preserving behavior.
- Done condition: Migration complete, auth tests pass, and no unrelated files change.
- Verification: Surface git diff summary plus the exact auth test command and passing output.
- Constraints: Auth-sensitive change — keep scope tight; do not weaken token validation or session handling without review.
- Execution strategy: Decide only after approval; keep security-sensitive integration under main-agent control and delegate only bounded work that inherits all auth constraints and receives independent review.
- Checkpoints: Report after each migration slice.
- Stop or ask when: Auth behavior would change in a way that needs review, or tests fail for an unclear reason.
- Prompt: none
- Next: ask approval
```
Auth, billing, production-data, schema, and other irreversible work stays behind explicit approval even at high fit. Do not call `create_goal` until the user approves.
## Codex tooling: active goal already present
User: "Set a new goal for this refactor."
`get_goal` shows another goal is active.
```text
Goal Gate
- Decision: defer
- Runtime: codex-tooling
- Goal fit: high
- Objective: Refactor the requested module while preserving behavior.
- Done condition: Refactor complete and validation passes.
- Verification: Surface active goal status, the proposed objective, and validation output once approved.
- Constraints: Do not replace or mutate the active goal without user approval.
- Execution strategy: n/a until the active-goal decision is resolved.
- Checkpoints: n/a
- Stop or ask when: A goal is already active.
- Prompt: none
- Next: ask approval
```
Ask whether to continue, complete, block, clear, or replace the active goal before any goal mutation.
## Claude Code slash: auto-adopt
User: "I'm in Claude Code. Implement docs/PLAN.md and keep going until `pnpm test` passes and the app builds. I'm stepping away — finish it and show me the evidence."
High fit, gate clear, but Claude Code has no goal API. So adopt the contract and keep working. The user did not ask for a reusable prompt, so `Prompt: none`; the block is the contract.
```text
Goal Gate
- Decision: set-now
- Runtime: claude-code-slash
- Goal fit: high
- Objective: Implement docs/PLAN.md.
- Done condition: All acceptance criteria hold, the app builds, and `pnpm test` passes.
- Verification: Surface acceptance-criteria mapping, test/build commands, exit codes, and screenshots of each UI acceptance item if UI changed.
- Constraints: Follow the plan; do not broaden scope; do not push.
- Execution strategy: Assess plan-task independence before implementation; use subagents only for bounded independently verifiable tasks, and keep review, conflict resolution, and final integration verification with the main agent.
- Checkpoints: Report after each acceptance-criteria group.
- Stop or ask when: The plan conflicts with code reality, a destructive operation is needed, or verification cannot run.
- Prompt: none
- Next: adopt goal and continue
```
Claude Code goal evaluators judge from surfaced conversation evidence, so the agent must print the commands and results it wants evaluated as it works.
## Cursor CLI: auto-adopt in the current chat
User: "I'm in Cursor CLI Agent mode. Implement docs/PLAN.md and keep going until `pnpm test` and `pnpm build` pass. Do not commit or push."
Cursor CLI has no verified product `/goal` or goal API. Explicit Cursor intent wins even when the selected model is Grok or Claude. High fit and a clear safety gate still use `set-now`, but activation is a soft adoption in the current chat:
```text
Goal Gate
- Decision: set-now
- Runtime: cursor-cli
- Goal fit: high
- Objective: Implement every in-scope task in docs/PLAN.md until the named validation commands pass.
- Done condition: Every PLAN.md acceptance criterion is implemented, `pnpm test` passes, and `pnpm build` succeeds.
- Verification: Surface the PLAN.md task map, changed files, commands, exit codes, and relevant logs.
- Constraints: Stay within PLAN.md scope; do not commit or push; do not claim product Goal activation.
- Execution strategy: Assess dependencies, shared state, write overlap, and independent verification before delegation; keep final integration with the main agent.
- Checkpoints: Report after plan inspection, each milestone, and final validation in this chat.
- Stop or ask when: The plan is missing or ambiguous, credentials or destructive work are required, or verification cannot run.
- Prompt: none
- Next: adopt goal and continue
```
Continue the work immediately. Do not emit `/goal`, ask the user to paste a prompt back, or call `get_goal`, `create_goal`, or `update_goal`. The contract is durable only through the Cursor chat transcript.
## Cursor CLI: resume the adopted contract
User resumes the same chat with an installed CLI option: `cursor-agent --resume [chatId]` selects a chat and uses the supplied ID when present, `cursor-agent --continue` or `cursor-agent resume` restores the most recent chat, and `cursor-agent ls` opens chat selection. Then they ask to continue the unfinished PLAN.md work.
```text
Goal Gate
- Decision: set-now
- Runtime: cursor-cli
- Goal fit: high
- Objective: Continue the exact PLAN.md contract already present in this chat.
- Done condition: The prior contract's full done condition is proven.
- Verification: Continue surfacing checkpoint and terminal evidence in this chat.
- Constraints: Preserve the prior scope and write boundaries; do not claim a product Goal is Active.
- Execution strategy: Continue the prior strategy; reassess only if code or task dependencies changed.
- Checkpoints: Continue the prior checkpoint cadence.
- Stop or ask when: The resumed transcript lacks the contract, the requested objective conflicts, or a prior pause condition fires.
- Prompt: none
- Next: continue adopted goal
```
Chat resume restores conversation context; it does not create or restore product goal state. Do not draft a second contract when the exact one is already in the resumed transcript.
## Cursor CLI: prompt only
User: "Write a Cursor CLI goal prompt for docs/design.md. I want to review it before running."
Use `Decision: suggest`, `Runtime: cursor-cli`, and `Next: provide prompt`. Provide a complete plain prompt with verification, constraints, boundaries, execution strategy, iteration, done, and pause conditions. Do **not** prefix it with `/goal`; Cursor's `/` skill invocation is not product goal activation.
## Claude Code slash: prompt only (user wants to review)
User: "Generate a Claude Code /goal prompt for implementing docs/design.md. I want to review it before running."
The user asked to review first, so hand back the prompt instead of adopting it.
```text
Goal Gate
- Decision: suggest
- Runtime: claude-code-slash
- Goal fit: high
- Objective: Implement docs/design.md.
- Done condition: All acceptance criteria hold, the app builds, and the specified tests pass.
- Verification: Surface acceptance-criteria mapping, test/build commands, exit codes, and screenshots of each UI acceptance item if UI changed.
- Constraints: Follow the design doc; do not broaden scope; do not push.
- Execution strategy: Assess whether the plan contains bounded independently verifiable tasks; delegate only when useful and keep the main agent responsible for synthesis and final verification.
- Checkpoints: Report after each acceptance-criteria group.
- Stop or ask when: The doc conflicts with code, a destructive operation is needed, or verification cannot run.
- Prompt: /goal Implement docs/design.md without stopping until every acceptance criterion is satisfied, the app builds, and the tests pass. Execution strategy: Before implementation, assess dependencies, shared context, write overlap, and independent verification to decide whether subagents help; keep the main agent responsible for synthesis and final verification. After each checkpoint, report what changed, what evidence was produced, what remains, and whether the done condition is met. Stop and ask if the doc conflicts with the code, verification is blocked, or a destructive action would be required.
- Next: provide prompt
```
## Codex slash: prompt for reuse
User: "Give me a Codex /goal prompt to improve `prompts/support-agent.md` until `pnpm eval:prompt -- --suite support-agent` reaches at least 0.92."
```text
Goal Gate
- Decision: suggest
- Runtime: codex-slash
- Goal fit: high
- Objective: Improve prompts/support-agent.md until the support-agent eval reaches at least 0.92.
- Done condition: `pnpm eval:prompt -- --suite support-agent` reaches 0.92, or further changes require product guidance.
- Verification: Surface each eval command, score, failing cases reviewed, and final diff summary.
- Constraints: Keep prompt edits minimal and do not change eval fixtures unless explicitly approved.
- Execution strategy: Delegate independent eval analysis only when it reduces context pressure; keep prompt edits serialized and final score/diff verification with the main agent.
- Checkpoints: Report after each eval run.
- Stop or ask when: The target cannot be reached without changing policy, product behavior, or eval fixtures.
- Prompt: see Recommended /goal below
- Next: provide prompt
```
```text
/goal Optimize prompts/support-agent.md until `pnpm eval:prompt -- --suite support-agent` reaches at least 0.92.
Verification: Run `pnpm eval:prompt -- --suite support-agent` after each focused change; surface the score, failing cases reviewed, and final diff summary.
Constraints: Keep prompt edits minimal; do not change eval fixtures, product behavior, or policy unless explicitly approved.
Boundaries: Write only prompts/support-agent.md; treat eval fixtures and runner code as read-only.
Execution strategy: Assess whether independent read-only eval analysis would reduce context pressure; delegate only when it does, serialize prompt edits, and keep final score/diff verification with the main agent.
Iteration policy: Run at most 5 focused edit/eval rounds; after two rounds with no score improvement, inspect failure clusters and change the hypothesis before retrying.
Stop when: The eval score reaches at least 0.92 with surfaced evidence, or no in-boundary prompt change can improve the remaining failures.
Pause if: Reaching the target requires changing policy, product behavior, eval fixtures, credentials, or another file outside the write boundary.
```
The user asked for a prompt to copy, so this stays `suggest` even at high fit.
## Unknown runtime
User: "I might run this in another coding agent. Make the goal portable."
```text
Goal Gate
- Decision: suggest
- Runtime: unknown
- Goal fit: medium
- Objective: Complete the requested refactor with behavior preserved.
- Done condition: Refactor is complete and the agreed validation signal passes.
- Verification: Surface changed files, validation command output, and behavior-parity notes.
- Constraints: No public API changes; no unrelated formatting churn.
- Execution strategy: Assess delegation only if the runtime supports it; otherwise use one agent, and in all cases keep scope review and final behavior-parity verification with the main agent.
- Checkpoints: Report after each logical slice.
- Stop or ask when: The runtime lacks required tools, validation is unavailable, or scope expands.
- Prompt: Complete the refactor until the agreed validation signal passes. Execution strategy: Assess whether delegation helps only when the runtime supports it; otherwise use one agent, while keeping final scope and behavior-parity verification with the main agent. Keep changes scoped, surface each verification result, and stop for ambiguity, unavailable tools, or scope expansion.
- Next: ask approval
```
Capabilities are uncertain, so do not auto-execute even if fit were high.
## Workflow gate already ran
Existing block says `Route: Architecture`, `Runtime skill: architecture-hardening-loop`, `scope=multi-module`, `user-intent=implement`, and `get_goal` shows no active goal.
Scoped `architecture-hardening-loop` plus implement intent is a strong goal-fit signal for a transcript contract, but it does not by itself authorize native Goal creation. With the safety gate clear, adopt and continue unless the user or parent workflow explicitly requested Goal state.
```text
Goal Gate
- Decision: suggest
- Runtime: codex-tooling
- Goal fit: high
- Objective: Complete the multi-module implementation selected by workflow-gate.
- Done condition: All planned modules are implemented and their validation commands pass.
- Verification: Surface per-module validation output and final git diff summary.
- Constraints: Preserve the workflow-gate route; do not broaden scope beyond the plan.
- Execution strategy: Assess bounded-context independence, dependency order, shared state, and write overlap; delegate or parallelize only non-conflicting work and keep aggregate verification with the main agent.
- Checkpoints: Report after each module or bounded context.
- Stop or ask when: The plan proves wrong, modules conflict, or validation is blocked.
- Prompt: none
- Next: adopt goal and continue
```
`workflow-gate` is a signal, not a prerequisite. Do not rerun it unless the user is asking for route selection rather than goal selection.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |