The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external sources such as pull requests, code repositories, and design documents. While it contains directives for structural skepticism and source-bound evidence, the broad ingestion surface makes it susceptible to malicious instructions embedded within the analyzed content.
Ingestion points: Pull requests, repositories, subsystems, and design documents specified during invocation (SKILL.md).
Boundary markers: The skill relies on architectural reconstruction and evidence-based reasoning rather than strict string delimiters for untrusted input.
Capability inventory: Execution of shell commands ('safe targeted checks') and the spawning of parallel investigator agents.
Sanitization: Mandates the sanitization of source excerpts and explicitly forbids publishing secrets or raw inherited tool payloads.
[DYNAMIC_EXECUTION]: The skill implements a complex multi-agent parallelization strategy described in 'prepared-review.md'. It uses platform capabilities like spawn_agent and 'native-fork' transports to create up to 20 parallel investigator agents with shared history checkpoints to analyze architectural obligations.
[COMMAND_EXECUTION]: The instructions authorize the use of 'safe targeted checks' and shell commands to gather evidence for the review. Although it requires the agent to inspect commands before execution and restricts external effects or production probes, this capability presents a standard local execution surface.

architectonic