Task Finisher with Grok
Finish multi-step tasks with per-step evidence and a publication-safety check before handoff.
Free
Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsTask Finisher with Grok
Example session with this skill installed
I need to [describe the task]. Apply Task Finisher with Grok and drive it to a complete, review-ready deliverable. If it is one step, use the trivial fast path.
- Read your context and instructions
- Compiled the task finisher
Filled Execution Summary for a two-step auth refactor.
## Execution Summary
**Original Request**
Type the token parser and cover the reject path with one test.
**Assumptions**
JWT stays the token format. No new login flow.
**Steps**
| Step | Criterion | Evidence |
| --- | --- | --- |
| 1. Type the token parser | `parseToken` returns `TokenClaims` | Re-read `src/auth.ts`: the return type is `TokenClaims` |
| 2. Cover the reject path | One test fails on a bad signature | `npm test -- auth.test` exit 0; "rejects a bad signature" passed |
**Final Deliverable**
Typed parser and the new test, in the auth module.
**Verification**
- All success criteria met: Yes
- Outstanding items: None
- Paths scrubbed: yes (0 replacements)
- Pre-existing files modified: None
**Recommended Next Steps**
- None
Connects securely to your tools. The creator never sees your data.
About this skill
Task Finisher with Grok drives stalled work to a review-ready deliverable. Each step has evidence. Machine-local paths, usernames, and hostnames are scrubbed before handoff.
One obvious step uses the Trivial fast path: do it, run the weakest check, run the Publication Safety gate, and hand off one row. Independent steps run together. Verification stays sequential.
Phase 1: Goal. Restate the request in one line and define done. Stop only when scope or success criteria are unclear. Otherwise assume and continue.
Phase 2: Breakdown. Split the work into steps. Independent steps share no inputs and no writes.
Phase 3: Execute and verify. Run independent steps in parallel. Write each evidence row before a later step depends on it.
Phase 4: Integrate and Publication Safety Check. Assemble the deliverable, compare it to the goal, and run the Publication Safety gate over everything that will be published, including files you did not write. warn (default) scrubs and delivers. off skips. strict blocks handoff if a match remains. ALLOW_LOCAL_PATHS true overrides strict.
Phase 5: Handoff. Present the deliverable and an Execution Summary with one row per step, listing every pre-existing file the scrub modified.
Use it for "finish this" or "I'm almost done". Not for brainstorming with no deliverable, designing a skill, converting a prompt, or auditing a skill.
- Open brainstorming with no deliverable is out of scope. So are designing a skill, converting a repeated prompt, and auditing an existing skill.
PUBLISH_SAFETY=strictstops handoff when a machine-local identifier remains. It does not delete the files to force a pass.PUBLISH_SAFETY=offandALLOW_LOCAL_PATHS=trueskip the scrub. The default is to scrub and deliver (warn, allow false).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
This skill has been used 13 times.
Trust & safety
Security scan
Security scan passed 3 days ago
- Free with an Agensi account
Permissions
File Scopes
If the task involves exporting content outside of the development environment or specific project workspace the agent must avoid leaking absolute local paths, usernames, or machine-specific information into the export target area. Environment variables are only required if it related to the specific project.