The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves auditing external project files which serves as a potential attack surface for indirect prompt injection.
Ingestion points: The scripts/audit_updater.py script reads multiple files from the target project directory, including package.json, Cargo.toml, tauri.conf.json, and various JavaScript/TypeScript/Rust source files.
Boundary markers: The skill instructions in SKILL.md use standard markdown headers to separate the audit output from other instructions, but there are no explicit delimiters or specific warnings to the agent to ignore instructions embedded within the audited code.
Capability inventory: The skill uses local scripts to perform file system operations (read/write) and instructions allow the agent to modify project source code based on audited findings.
Sanitization: scripts/audit_updater.py employs file suffix filtering (.js, .ts, .rs, etc.) and directory exclusion (e.g., node_modules, .git) to limit the scope of ingested data.
[COMMAND_EXECUTION]: The skill provides and instructs the agent to execute several local Python scripts (audit_updater.py, profile_store.py, validate_skill.py) to perform its core functions.
Evidence: README.md and SKILL.md provide specific command-line examples for running these scripts, such as python3 scripts/audit_updater.py /path/to/project. While these scripts perform legitimate auditing and configuration tasks, the execution of local scripts is a privileged capability that requires oversight.

lov-app-delta-updater