The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates a temporary Playwright test file (tests/e2e/.tmp/selector-probe.spec.ts) during Phase 5 to verify if a new selector resolves against the live application. This generated script is then executed using the playwright test command. This represents runtime code generation and execution based on analyzed evidence. \n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from Dash0 telemetry spans and Playwright trace artifacts, including HTML snapshots, console logs, and network data. This information is used to synthesize root-cause hypotheses and draft automated code fixes, creating a vulnerability surface where a malicious test environment or telemetry could influence the generated code. \n
Ingestion points: Dash0 telemetry spans (Phase 1) and Playwright trace.zip files (Phase 2) as documented in rules/telemetry-driven-analysis.md and rules/local-iteration.md. \n
Boundary markers: The skill does not define specific boundary markers for the LLM when processing this external data. \n
Capability inventory: The skill has access to shell execution (Bash), file system modification (Edit, Write), and remote repository interaction (git push). \n
Sanitization: The skill documents a "privacy pre-flight" in rules/self-improvement-loop.md to remove PII and credentials before persisting data to its memory loop. \n- [EXTERNAL_DOWNLOADS]: The skill downloads tooling and artifacts from external sources, including npx playwright init-agents and gh run download. These operations target trusted services (Microsoft and GitHub) and are required for the skill's primary functionality. \n- [COMMAND_EXECUTION]: The skill relies on various CLI tools (git, gh, node, pnpm, rg, jq) to perform repository management, telemetry querying, and local test execution as part of its automated stabilization workflow.

e2e-pr-stabilizer