PR + Commit Workflow

This skill should be used when creating commits or pull requests, enforcing a human-written PR structure, intent capture, and evidence in agentic workflows.

This skill aims to produce high-quality, auditable PRs, but it requires collecting and embedding 'full prompt history' and environment metadata into PR bodies — data that often contains sensitive info (API keys, passwords, private prompts). Before installing, consider: 1) Do you ever store secrets or private data in prompts or agent logs? If yes, do NOT enable automatic inclusion of full prompt history. 2) The skill expects to run gh/git commands but doesn't declare those binaries — ensure gh is installed and that you trust the skill to run it. 3) The provided scripts read many environment variables and look for local agent files; review scripts/build_pr_body.sh locally to see exactly what will be collected. 4) Ask the maintainer (or modify locally) to: a) require explicit user consent before collecting/publishing prompt history or environment metadata; b) make redaction rules strict and automatic (remove credentials, tokens, secrets) rather than leaving redaction to the agent's discretion; c) add an option to include only metadata (harness/model) rather than full prompts; d) declare required binaries (gh, git) in metadata. 5) If you want to use this skill but restrict risk: disable autonomous invocation for it, require a human confirmation step before any collection/publish, and test the build_pr_body.sh script in a sandbox to confirm it doesn't reveal anything you consider private. If these mitigations cannot be implemented, treat the skill as unsafe for repositories where prompt history or environment metadata may include secrets.

Type: OpenClaw Skill Name: pr-commit-workflow Version: 1.0.0 The skill is classified as suspicious due to the execution of a shell script (`scripts/build_pr_body.sh`) that reads various environment variables and system information, and the instruction in `references/workflow-commit.md` to potentially execute an arbitrary local 'committer script' if present in the repository. While these actions are plausibly aligned with the stated purpose of collecting environment metadata for auditability and adapting to repo-specific tooling, they represent high-risk capabilities (shell execution, reading environment variables, executing external scripts) without clear malicious intent, thus warranting a 'suspicious' classification.