The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files to plan and execute a QA matrix, which creates a vulnerability surface for instructions embedded in data.
Ingestion points: The skill reads project documentation including _tasks.md, _prd.md, task_NN.md, and Surface Transcripts (File: SKILL.md, Section 1 and 2).
Boundary markers: The skill does not specify the use of delimiters or instructions to ignore embedded commands within the ingested documentation.
Capability inventory: The agent is instructed to use shell commands (make build, git diff-tree, roundfix), access the file system for report generation, and perform network requests for API and frontend testing.
Sanitization: No explicit sanitization or filtering of content from the PRD or Task files is mentioned before the agent uses that content to formulate testing steps.
[COMMAND_EXECUTION]: The instructions direct the agent to execute specific shell commands to audit, build, and verify the project state.
Evidence: Section "Non-negotiables" (point 6) requires the use of make build and ./bin/roundfix. Section "Surface Transcripts" specifies running roundfix spec check <slug> --strict. Section "Tooling audit" utilizes git diff-tree --no-commit-id --name-only -r <commit>.
[DYNAMIC_EXECUTION]: The skill facilitates the runtime compilation and execution of project-specific binaries to perform testing flows.
Evidence: In Section "Non-negotiables" (point 6), the agent is instructed to build a binary from the current worktree using make build and subsequently execute that generated binary (./bin/roundfix) to perform CLI user flows.

qa-gate