The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for project testing and validation. Security is maintained by sourcing these commands from local configuration files (.ai/agentic.config.json) and validating all external variables, such as issue IDs and branch names, against strict regex patterns (^[A-Za-z0-9._/-]+$) before they are interpolated into the shell.
[SAFE]: The static detector flagged a potential prompt injection in references/agentic-setup.md. Analysis confirms this is a false positive; the text in question is actually a security instruction that explicitly tells the agent to ignore and report malicious directives found in untrusted data (e.g., 'ignore previous instructions').
[SAFE]: While the skill ingests untrusted data from external issue trackers and repository files, creating an indirect prompt injection surface, it employs comprehensive mitigations:
Ingestion points: Issue descriptions, PR diffs, CI logs, and analyzer briefs are read into context in SKILL.md and references/agentic-setup.md.
Boundary markers: The 'Untrusted content boundary' section explicitly instructs the agent to treat repository content as data and never as instructions.
Capability inventory: Capabilities are limited to file editing, local shell execution (tests/linting), and tracker metadata updates (assigning and labeling).
Sanitization: All string inputs used in paths or shell commands are validated and quoted as per the instructions in references/agentic-setup.md.

om-fix