The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill employs the !command syntax to execute a shell command during the skill's load phase. The command bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config ... is used to fetch project-specific configurations. This is a legitimate use case for project-specific tooling and configuration resolution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from potentially untrusted files in the repository, creating a surface for indirect prompt injection.
Ingestion points: Accesses CLAUDE.md, files in production/milestones/ and production/releases/, and design/live-ops/content-calendar.md. It also performs broad codebase scans via Glob, Grep, and Read tools to identify TODOs, debug logs, and placeholder assets.
Boundary markers: The skill does not implement delimiters or safety instructions to prevent the model from obeying instructions potentially embedded in the files it reads.
Capability inventory: The skill possesses Write access to the filesystem (specifically for saving the checklist) and a restricted Bash tool limited to a single configuration helper script.
Sanitization: There is no evidence of sanitization or input validation for the data ingested from the repository files.

launch-checklist