The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The AI code review workflow defined in assets/claude-code-review.yml and assets/copilot-review-instructions.md is designed to ingest and process untrusted data from pull request diffs, metadata, and user comments.
Ingestion points: Untrusted data enters the agent context via gh pr diff, gh pr view, and gh pr list commands in the claude-code-review.yml jobs.
Boundary markers: The prompts do not use specific XML tags or robust delimiters to wrap the output of the gh commands, which could allow malicious content in a PR to influence the agent's behavior.
Capability inventory: The workflow grants the agent capabilities to post inline comments, create PR-level comments, and read workflow logs across multiple parallel jobs.
Sanitization: There is no explicit sanitization or filtering of the PR diff content before it is processed by the AI agent.
[EXTERNAL_DOWNLOADS]: The skill relies on several external utilities and scripts to perform its automation tasks.
Package installation: The skill metadata and review workflows install the skills package via NPM and require goreleaser and gh via Homebrew.
Remote script execution: The claude-code-review.yml workflow downloads additional Go-specific agent skills from https://github.com/samber/cc-skills-golang using npx skills add at runtime.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage the development environment and CI pipelines.
The agent is explicitly allowed to use Bash(go:*), Bash(git:*), Bash(goreleaser:*), and Bash(gh:*) as defined in the allowed-tools frontmatter.
The dependabot-auto-merge.yml asset executes gh pr merge to modify the repository state, though it is protected by a github.actor guard.

golang-continuous-integration