The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on project-local configuration and profile files to determine its operational parameters, accompaniment levels, and autonomous behavior.
Ingestion points: The skill ingests data from 02-DOCS/wiki/harness/user-profile.md to set the 'accompaniment dial' and from 02-DOCS/wiki/sdd/config.yaml to configure 'Autopilot' mode and model routing.
Boundary markers: There are no explicit delimiters or 'ignore embedded instructions' markers defined for the data read from these configuration files.
Capability inventory: The skill has the ability to write multiple artifacts to the 02-DOCS/wiki/sdd/ directory, invoke sibling phase skills (e.g., ../specify/SKILL.md), delegate tasks via the Task tool, and issue commands to the assistant host.
Sanitization: The skill instructions do not specify any validation, escaping, or filtering for the external configuration content prior to use in execution logic.
[COMMAND_EXECUTION]: The skill directs the agent to issue platform-specific control commands, such as /model <name>, to dynamically switch the active language model for the session or for specific sub-tasks. This capability is used to optimize cost and reasoning quality across different development phases.
[DYNAMIC_EXECUTION]: The skill implements an 'Autopilot' mode which, upon initial user consent, enables the agent to execute a multi-step chain of development phases (from specification through to verification and review) without requiring intermediate human approval between steps. This represents a high degree of agent autonomy in the development process.

sdd