The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/e2e-process-project.js executes shell commands using execFileSync with the shell: true option enabled. This configuration is vulnerable to command injection because arguments such as the project name (project) and repository (repo) are interpolated into the shell command. The project variable is derived from GitHub artifact names in scripts/e2e-gather-run-info.js. If an attacker can control artifact names in the analyzed repository, they can execute arbitrary shell commands on the system where the skill is running.
[REMOTE_CODE_EXECUTION]: The skill downloads and processes external artifacts and reports from GitHub and S3/CloudFront. It executes npx playwright merge-reports and other tools on this downloaded content. While intended for analysis, this creates a path for executing logic based on remote, untrusted data.
[PROMPT_INJECTION]: The skill has a significant Indirect Prompt Injection surface as it is designed to ingest and summarize large amounts of untrusted data from CI environments.
Ingestion Points: CI job logs, Playwright JSON reports, and accessibility-tree snapshots are downloaded and read by the agent across several scripts including scripts/e2e-process-project.js and scripts/e2e-process-s3.js.
Capability Inventory: The agent has access to the local file system, the gh CLI (often authenticated with high-privilege tokens), and network capabilities through Node.js scripts.
Sanitization: The skill does not implement sanitization or boundary markers to prevent malicious instructions embedded in the logs or reports from influencing agent behavior.
[DATA_EXFILTRATION]: The script scripts/e2e-query-history.js transmits the E2E_INSIGHTS_API_KEY and test data to https://connect.posit.it/e2e-test-insights-api. This is a legitimate connection to a vendor-provided service for the author 'posit-dev'.

e2e-failure-analyzer