The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file contains explicit instructions for the agent to run shell commands to install external skills (e.g., python3 scripts/install-external-skills.py --skill browserbase-search --agent codex). This encourages the execution of scripts that are not provided within the skill package.
[REMOTE_CODE_EXECUTION]: The skill references an external registry (../../../references/external-skills.yaml) and provides commands to install code from this registry, which could result in the execution of unverified remote code if the registry or the installation script is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, including customer interview transcripts and prospect emails.
Ingestion points: Transcripts are processed by scripts/customer_interview_analyzer.py (line 48); inbound leads and emails are processed as described in references/prospect-enrichment-and-qualification.md.
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" warnings for the processed data.
Capability inventory: The skill includes Python scripts for analysis and instructions in SKILL.md to execute shell commands for skill installation.
Sanitization: There is no evidence of input sanitization or validation in the provided scripts to prevent malicious instructions embedded in data from affecting agent behavior.

discovery