The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external blueprints and implementation details which are untrusted sources of data. An attacker could potentially embed malicious instructions within a blueprint to influence the agent's behavior during the test generation or execution phases.
Ingestion points: The skill reads implementation blueprints (SKILL.md), repository revisions, and service context (references/input-and-report.md).
Boundary markers: The skill records input snapshots and content hashes to detect unauthorized changes to the blueprint, although it lacks explicit delimiters for processing the blueprint's natural language content.
Capability inventory: The agent can execute arbitrary shell commands for test runners, generate and execute Playwright scripts, and write files to a designated artifact destination (SKILL.md, references/tool-selection.md).
Sanitization: The instructions include explicit requirements to protect credentials, use secret placeholders, and redact sensitive fixture data (references/input-and-report.md, references/playwright.md).
[COMMAND_EXECUTION]: The skill executes test suites and commands defined in the repository or inferred from the blueprint.
Evidence: Procedure Step 4 in SKILL.md requires the agent to "Run the mapped unit/integration suites and affected regressions" and save the commands and exit statuses.
[DYNAMIC_EXECUTION]: The skill generates new executable test artifacts (e.g., Playwright tests) based on the input blueprint and subsequently executes them.
Evidence: references/test-generation.md describes the process of generating missing acceptance checks and auditing assertions before execution. references/playwright.md describes using Playwright to convert a plan into executable tests. The skill correctly references official documentation from Playwright, a well-known service.

qa