The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to "Run every command in the Task's ## Verification section verbatim." This design allows for the execution of arbitrary shell commands parsed directly from documentation files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from various specification files which dictate the agent's behavior and command execution.
Ingestion points: Markdown files located in docs/specs/<slug>/ (including task_NN.md, _prd.md, and _techspec.md).
Boundary markers: No delimiters or safety warnings are provided to prevent the agent from obeying instructions embedded within these files.
Capability inventory: The agent is authorized to execute shell commands, perform file system mutations, and engage in "Characterization Tasks" that reach external boundaries using credentials.
Sanitization: No sanitization or validation of the extracted commands is performed; the instructions explicitly require "verbatim" execution.
[PROMPT_INJECTION]: The instructions contain directives to override standard agent safety and interaction patterns, such as "Begin immediately: no greeting, no plan-approval question, no waiting for a 'go'." and "Authorization to start is implied by assignment."

implement-task