The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local shell scripts (e.g., scripts/pre-tag-validate.sh) and standard version control commands (git tag, git push) to validate the project state and perform release operations. These actions are aligned with the skill's primary function as a release conductor.
[PROMPT_INJECTION]: The skill uses an indirect prompt injection surface by design, specifically through a "reference + execute inline" pattern where it ingests instructions from separate local files at runtime.
Ingestion points: The skill reads subagents/pm-skill-auditor.md and subagents/pm-changelog-curator.md to define its behavior at specific release gates.
Boundary markers: Absent; the skill directs the agent to adopt the file content as its operating instructions without explicit delimiters to isolate the injected content.
Capability inventory: The agent has access to Bash, Edit, Read, Grep, Glob, and Agent tools across the conductor and its inlined children.
Sanitization: No filtering or sanitization of the referenced markdown instructions is performed before they are adopted into the execution context.
[COMMAND_EXECUTION]: The skill dynamically executes instructions loaded from external markdown files, effectively updating its behavioral logic during the release flow based on the contents of the subagents/ directory.

utility-pm-release-conductor