The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a fallback procedure to fetch its manifest documentation and rendering script from the author's GitHub repository (https://raw.githubusercontent.com/christopher-kapic/skills/master/skills/engineering/explore-offload/) if the local files are inaccessible.
[REMOTE_CODE_EXECUTION]: If the fallback mechanism is triggered, the skill executes the downloaded render.sh script using the shell (sh).
[COMMAND_EXECUTION]: The skill executes a local shell script (scripts/render.sh) to bundle repository excerpts. Additionally, the procedure involves executing user-provided shell commands (e.g., { ; } > /.log) to generate test logs for analysis.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a subagent to process repository files and logs, creating an attack surface for indirect prompt injection from untrusted source code or log data.
Ingestion points: The scout subagent reads files and log outputs from the repository as instructed in SKILL.md and references/manifest.md.
Boundary markers: No explicit boundary markers or directives to ignore instructions within ingested data are implemented.
Capability inventory: The skill has the ability to execute shell and git commands, and read/write files in the scratch directory.
Sanitization: The rendering script filters binary files and git-ignored paths, but the skill does not perform content-level sanitization on the data ingested by the subagent.

explore-offload