The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands to manage the release process, including git, npm, and gh. It specifically instructs the agent to navigate to ~/Scripts/claude-mem/ to execute a Discord notification script, which targets a path outside the immediate repository context.
[DATA_EXFILTRATION]: Accessing ~/Scripts/claude-mem/ to utilize a .env file containing Discord webhook details constitutes access to sensitive configuration data. While the operation is intended for notifications, it demonstrates the ability to read and use credentials stored in the user's home directory.
[PROMPT_INJECTION]: The skill uses ! [command] style instructional markers like **CRITICAL:** and **IMPORTANT:** to direct agent behavior. While these are currently used for workflow integrity, they mirror patterns used in prompt injection to override default AI behavior.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface: The scripts/generate_changelog.js script processes JSON data typically fetched from the GitHub API (gh release data). This data (release bodies/notes) is generated by users and then written directly into CHANGELOG.md.
Ingestion points: scripts/generate_changelog.js reads from stdin (Step 9 in SKILL.md).
Boundary markers: None identified; the script parses the JSON and joins the body text directly into a Markdown file.
Capability inventory: The skill has git commit and git push capabilities (Step 10 in SKILL.md), allowing it to persist modified files to the repository.
Sanitization: There is no evidence of sanitization or escaping of the r.body content before it is included in the changelog output.

version-bump