The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves reading externally provided QA artifacts and bug reports to generate automated code fixes using the file-edit tool.
Ingestion points: Reads various workspace files including story.*.qa.*.md, story.*.gate.*.yml, story.*.bug.*.md, task.*.qa.*.md, task.*.gate.*.yml, and task.*.bug.*.md (processed in SKILL.md Steps 1a and 1b).
Boundary markers: Employs code fences and clear behavioral directives in the subagent prompt (references/qa-findings-ingester-prompt.md) to maintain data/instruction separation.
Capability inventory: Possesses high-privilege capabilities including modifying source code via file-edit, executing build and test commands (nx, deno, git), and performing network operations via GitHub, Bitbucket, and Jira APIs.
Sanitization: Uses specialized local parsing logic (references/yaml-subset.js, references/change-log.js) and the Python SafeLoader (references/read-config.sh) to interpret structured data from potentially untrusted artifacts safely.
[COMMAND_EXECUTION]: Orchestrates various CLI utilities to perform development tasks, platform interactions, and environment discovery. Tools used include git, gh, curl, node, nx, and deno (referenced in SKILL.md and helper scripts like references/resolve-platform.sh).
[DYNAMIC_EXECUTION]: The configuration reader (references/read-config.sh) generates a Python script as a shell variable and executes it at runtime via python3 -c. While the Python source logic is a static template within the script, this pattern constitutes dynamic execution of logic derived from shell variables.

qa-fix