The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bundled bash scripts (scripts/sync_repo.sh and scripts/poll_analysis.sh) to perform operations such as git bundle create and git apply. These scripts interact directly with the local file system and repository state.
[DATA_EXFILTRATION]: The skill transmits repository content to the Sentinel service at sentinel.deepsource.com or a signed upload URL provided by the service for analysis. This involves sending source code and git history to an external endpoint as part of its core functionality.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing external data to modify local files.
Ingestion points: Analysis results and unified diff patches are fetched from the Sentinel API (GET /analysis/{id}) in SKILL.md and scripts/poll_analysis.sh.
Boundary markers: No delimiters or instructions to ignore embedded content are used when handling patches.
Capability inventory: The skill enables the agent to modify the filesystem via git apply and communicate over the network via curl.
Sanitization: The skill does not validate or sanitize patches received from the remote API before they are applied to the codebase.
[DYNAMIC_EXECUTION]: The shell scripts use python3 -c to execute Python code for parsing JSON data returned by the API. This involves execution of strings as code to handle data processing tasks.

sentinel-api