The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to perform shallow clones (git clone --depth 1) of external, third-party repositories from GitHub (GPTomics/bioSkills and ClawBio/ClawBio) to a local cache directory (~/.hermes/cache/scratch/).
[REMOTE_CODE_EXECUTION]: The skill explicitly states that the ClawBio repository contains "runnable pipeline skills" and executable Python scripts. It instructs the agent to run these scripts, specifically mentioning that many have "--demo flags and can be run directly."
[PRIVILEGE_ESCALATION]: The instructions include the use of sudo apt install to install bioinformatics tools (samtools, bcftools, etc.) on Linux platforms, which grants the skill/agent elevated administrative permissions.
[COMMAND_EXECUTION]: The skill provides complex environment setup instructions involving multiple package managers (pip, conda, apt, brew) and language-specific installers (BiocManager for R). This allows for broad system modifications and the execution of external installers.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and follow instructions from external repositories to perform tasks.
Ingestion points: The agent is instructed to cat and read content from SKILL.md and README.md files downloaded from the GPTomics and ClawBio GitHub repositories.
Boundary markers: There are no instructions provided to the agent to treat this external content as untrusted or to use delimiters to separate these external instructions from its primary system prompt.
Capability inventory: The agent has the capability to execute shell commands, run Python scripts, and install system-level packages based on the guidance found in these external files.
Sanitization: No sanitization or validation of the fetched external markdown content is performed before the agent follows it as "reference material."

bioinformatics