The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing a third-party extension from an untrusted GitHub repository to enable specific messaging capabilities between different agent harnesses.
Evidence: Step 3 of the 'Harness setup' in SKILL.md suggests running pi install git:github.com/alonw0/pi-claude-link if the extension is missing from the user's environment.
[INDIRECT_PROMPT_INJECTION]: The skill manages a multi-agent workflow that processes external task descriptions, creating a surface for indirect injection attacks where malicious task content could influence the leader or sub-agents.
Ingestion points: Ingests task data from issue trackers and user messages as defined in 'Workflow' step 1 of SKILL.md.
Boundary markers: While the skill uses a structured briefing template, it lacks explicit delimiters or instructions to ignore embedded commands within the external task content.
Capability inventory: The skill utilizes powerful tools including shell execution, remote terminal management (orca, herdr), and remote file system access via ssh (e.g., in retire-agent.sh).
Sanitization: Employs printf %q for shell argument escaping and jq for JSON manipulation, which provides significant protection against direct command injection during orchestration.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses platform-specific syntax to execute shell commands at load time to display its own configuration to the agent.
Evidence: SKILL.md contains the pattern !cat "${XDG_CONFIG_HOME:-$HOME/.config}/hyper/team-lead/harnesses.yaml" ... which executes to load existing harness settings when the skill is loaded by the agent harness.

team-lead