The Agent Skills Directory

[PROMPT_INJECTION]: The skill functions as an orchestrator that ingests data from the client-intake skill, specifically within the project requirements and client context fields. This data is subsequently passed to downstream implementation skills (e.g., project-management, lead-dev), creating a surface for indirect prompt injection. If user-provided requirements contain malicious instructions, they could potentially influence the behavior of agents receiving the dispatched tasks.
Ingestion points: agents/execution/task-dispatcher.md (accessing task.context.requirements).
Boundary markers: The provided logic does not explicitly define markers or delimiters to isolate untrusted user data from agent instructions.
Capability inventory: The orchestrator itself has limited permissions, but it dispatches work to implementation skills that may have file system or network access.
Sanitization: There is no evidence of string sanitization or instruction-filtering for ingested data within the orchestration scripts.
[COMMAND_EXECUTION]: The skill contains a suite of test scripts in the tests/ directory designed for development-time validation. These scripts utilize Node.js and shell commands to read files and verify the skill's directory structure.
Evidence: tests/run-tests.sh and tests/utils.js utilize the Node.js fs module to scan the agents directory.
[SAFE]: All components, including the state machine, queue manager, and audit logger, operate as described in the documentation. No obfuscated code, hardcoded credentials, or unauthorized external communications were detected.

task-orchestrator