The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: Accesses potentially sensitive files like .env and scans Git history to detect leaked secrets before a public launch. This is a defensive feature that explicitly instructs the agent to report only the file location and not the sensitive values themselves.
[COMMAND_EXECUTION]: Utilizes git and gh CLI tools to manage repository state, visibility, and history, which are standard for the skill's stated purpose.
[PRIVILEGE_ESCALATION]: Performs git push --force-with-lease during the history rewrite stage. While this is a high-permission operation, it is restricted to a specific optional stage, requires explicit multi-step confirmation, and uses the safer --force-with-lease variant.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted repository content to generate launch marketing materials.
Ingestion points: Reads project manifests (package.json, pyproject.toml, etc.), source code comments, and existing documentation from the target repository.
Boundary markers: Does not explicitly define delimiters for untrusted content but includes strict instructions to verify all generated claims against the codebase.
Capability inventory: Possesses Write, Edit, and Bash capabilities to modify the repository and push to GitHub.
Sanitization: Relies on mandatory human-in-the-loop confirmation before any files are edited or visibility is changed.
[DYNAMIC_EXECUTION]: Generates temporary configuration files (replacements.txt) and backup bundles to facilitate Git history rewriting using git filter-repo.

oss-launch