The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources during the dependency analysis phase.
Ingestion points: External GitHub repository release notes and CHANGELOG files are processed by the package-upgrade-analyzer sub-agent in SKILL.md (Workflow Step 4).
Boundary markers: The skill lacks explicit boundary markers or "ignore instructions" warnings when processing the output of the sub-agent.
Capability inventory: The agent has the capability to write to the file system (.claude/scratchpad/PR.md), execute shell commands (git, gh pr create, pnpm, uv), and delete files (rm).
Sanitization: No sanitization or validation of the external content is performed before the agent incorporates the information into its strategic migration plan.
[INDIRECT_PROMPT_INJECTION]: User-supplied arguments are directly interpolated into the skill's instructions.
Ingestion points: The $ARGUMENTS placeholder is located at the end of SKILL.md within a <user-guidelines> block.
Boundary markers: While wrapped in XML-style tags, these are treated as behavior overrides, providing a vector for user-supplied instructions to bypass previous constraints.
Capability inventory: The skill utilizes potentially sensitive tools like gh for pull request creation.
Sanitization: None.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch and analyze information from external websites.
Evidence: The workflow and PR template direct the agent to retrieve documentation and release metadata from github.com repositories.
[COMMAND_EXECUTION]: The skill performs various shell operations to manage dependencies and version control.
Evidence: Usage of pnpm, uv, git, and gh tools are integrated into the workflow for checking outdated packages, updating them, and creating pull requests.

bump-deps