The Agent Skills Directory

[PROMPT_INJECTION]: The skill's core logic is driven by external configuration files and source code, creating a surface for indirect prompt injection where malicious content in those files could influence the automated fix loop.
Ingestion points: Reads instructions and test plans from .context/qa-map.json, .context/plan-decisions.json, and project source files (read in Step 2).
Boundary markers: Absent. The skill does not implement delimiters or explicit instructions to the model to ignore potential overrides embedded in the data it processes.
Capability inventory: The skill uses high-impact tools including Write, Edit, and Bash to modify source code and execute commands based on the ingested data.
Sanitization: No evidence of input validation or sanitization is present for data read from files before it is used to determine code changes or command arguments.
[COMMAND_EXECUTION]: The skill uses the Bash tool to interact with the system environment and version control.
Evidence: Executes git diff to determine scope and runs shell scripts to verify the presence of testing utilities.
[EXTERNAL_DOWNLOADS]: The skill attempts to locate and execute a local browser testing utility, which is treated as an external dependency relative to this skill's own code.
Evidence: Searches for an executable located at .claude/skills/browse/browse/dist/browse or ~/.claude/skills/browse/browse/dist/browse to perform live testing.

qa