The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses gitpod environment ssh across multiple stages (02-05) to execute arbitrary shell commands in remote environments, including service management (pkill), git identity modification, and build/test tool execution.\n- [REMOTE_CODE_EXECUTION]: In Stage 05, the skill invokes the claude CLI on a remote Gitpod environment with the --dangerously-skip-permissions flag. This allows a sub-agent to autonomously generate and execute code (writing and running tests) without user oversight or approval.\n- [CREDENTIALS_UNSAFE]: The skill accesses sensitive platform environment variables, specifically PYLOT_DISPATCH_TOKEN for API authentication with PYLOT_API and $GITCONFIG (extracted from platform secrets) to configure git identities.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).\n
Ingestion points: Pull request metadata (titles, branches) and repository files (diffs, package.json, Gemfile, requirements.txt) are fetched via gh and git CLI tools.\n
Boundary markers: No delimiters or protective instructions are provided to sub-agents or the claude -p invocation when processing untrusted PR data.\n
Capability inventory: The skill possesses extensive capabilities, including remote shell access, package installation, Docker builds, and autonomous agent triggers.\n
Sanitization: There is no evidence of validation or sanitization of external data before it is passed into execution contexts.

deps-runner