The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends cloning and installing external AI agent skills from github.com/ChristopherKahler/seed and github.com/ChristopherKahler/paul. These repositories originate from an untrusted source and are recommended for installation into the local agent skill directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes screen-recording transcripts in tasks/from-recording.md and audits existing SOPs in tasks/audit.md. These inputs are untrusted external data and lack explicit boundary markers or sanitization, potentially allowing embedded malicious instructions to influence agent behavior.
Ingestion points: Screen-recording transcripts in tasks/from-recording.md and existing SOP files in tasks/audit.md.
Boundary markers: None identified in the tasks processing external content.
Capability inventory: The skill has Write and Bash access as defined in SKILL.md.
Sanitization: No content sanitization or validation is performed on ingested transcripts or audited SOP files.
[COMMAND_EXECUTION]: The skill uses the Bash tool for local environment reconnaissance and filesystem management.
Evidence: In tasks/interview.md and frameworks/automation-flagging.md, the skill executes shell commands to check for the presence of other skills in ~/.claude/skills/.
Evidence: In tasks/scaffold-business.md and tasks/interview.md, it uses mkdir and git rev-parse to organize the SOP directory structure.

sop-build