The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub pull request descriptions to automate changelog generation.
Ingestion points: PR descriptions are fetched via GraphQL and processed by the references/changelog-lines.jq script to extract user-facing changes.
Boundary markers: The processing script uses regex patterns to identify the ## Changelog section within PR bodies to isolate data.
Capability inventory: The skill can execute local shell scripts (./scripts/bump-version.sh, ./scripts/release-pretag-guard.sh), perform git tagging/pushing, and trigger GitHub workflows via the gh CLI.
Sanitization: The jq script performs basic text cleaning, such as removing HTML comments and normalizing line endings, but does not provide specific protection against malicious instructions embedded in the PR text that could target the LLM during the classification or release phase.
[COMMAND_EXECUTION]: The skill relies on several local shell scripts and external CLI tools for its core release functionality.
Evidence: Execution of ./scripts/bump-version.sh, ./scripts/release-pretag-guard.sh, and the glaeda-gh tool for release synchronization is documented in SKILL.md.
Context: These scripts are part of the cmux repository and are used for versioning, project metadata updates, and CI workflow management.

cmux-release