Task Finish Contract

Enforce task completion with explicit Goal/Progress/Next state. Prevent mid-task stalls and ensure evidence-based completion.

This skill is an instruction-only policy to make agents finish tasks and produce evidence; it does not install code or ask for credentials. Before enabling it, ensure the agent's workspace and memory files do not contain secrets (the skill expects to read/write memory/{date}.md and artifact paths). Note the example verification commands use PowerShell (Select-String) — if your agent runs on a non-Windows environment, adapt those checks (e.g., grep). Be aware that the skill encourages autonomous follow-up steps; if you want to limit any automated actions, control the agent's invocation policies or test the skill in a sandboxed workspace first.

Type: OpenClaw Skill Name: task-finish-contract Version: 1.0.0 The skill bundle is classified as suspicious due to the presence of executable PowerShell commands within the `SKILL.md` and `references/privacy-checklist.md` files. While these commands are intended for benign purposes (task completion verification and privacy scanning), their inclusion in markdown constitutes a potential prompt injection vector. If an AI agent is instructed to execute these commands, and if arguments (such as file paths) can be influenced by an attacker, it could lead to shell injection vulnerabilities. However, there is no evidence of malicious intent, such as data exfiltration or unauthorized remote control; in fact, `SKILL.md` includes explicit privacy and safety guidelines.