The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override interactive safety behavior by suppressing confirmation prompts. It directs the agent to "Do not ask for confirmation," "do not wait for further instructions," and "start executing it immediately," characterizing the initial invocation as total authorization for the entire automated loop.
[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and parsing several markdown files from the local workspace which may contain untrusted data that could influence the agent's behavior.
Ingestion points: The skill reads _prd.md, _techspec.md, _tasks.md, and various task_NN.md files to resolve the implementation graph.
Boundary markers: There are no specific delimiters or instructions defined to ensure the agent ignores or sanitizes instructions that might be embedded within these markdown files.
Capability inventory: The skill triggers a sequence of high-privilege operations, including code implementation (implement-task), automated testing (qa-gate), and file archival (archive-spec).
Sanitization: The skill lacks validation or filtering mechanisms for the content ingested from the external specification files before it is processed by the loop.

implement-spec