The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute local scripts for workflow orchestration, such as link-run.sh, scaffold-phase-deliverables.sh, and statusline.sh. It also executes external testing tools like npx playwright. There is a potential risk of command injection if parameters like RUN_ID, which are derived from user-provided task descriptions, are not properly sanitized before being passed to shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Jira tickets (via .claude/logs/jira/{TICKET_ID}.json) and user-provided task descriptions. This content is used as a canonical source for requirements, creating a vulnerability where malicious instructions embedded in tickets or tasks could be processed by the agent.
Ingestion points: Jira ticket log files and user task descriptions.
Boundary markers: No specific delimiters or "ignore instructions" markers for external data are defined.
Capability inventory: Access to Bash, Write, Edit, and subagent dispatch capabilities.
Sanitization: No explicit sanitization or validation of the external content is described.
[DYNAMIC_EXECUTION]: The skill uses the Agent tool to dynamically dispatch subagents (e.g., architect, tester, security) based on the current phase of the workflow. The agent identifiers are constructed at runtime using namespaced prefixes, representing dynamic execution flow orchestration.

run-orchestrator