The Agent Skills Directory

[DATA_EXPOSURE_EXFILTRATION]: The skill documents a workflow for capturing network traffic (HAR) which contains sensitive information such as Bearer tokens, cookies, and prompt data. It proactively mitigates this risk by instructing the user to store these in gitignored directories (scripts/dev/_spike_out/) and use a specific redaction script (extract_har_summary.py) before sharing any output to prevent accidental data exfiltration.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to probe and read the DOM and network logs of external websites (e.g., flow.google.com), the skill operates on untrusted external data. This creates a potential surface for indirect prompt injection if the resulting data were to be processed by an AI agent without sanitization.
Ingestion points: DOM, network traffic logs, and HAR files captured in scripts/dev/_spike_out/.
Boundary markers: None explicitly defined in the text, as the focus is on manual developer investigation.
Capability inventory: Local script execution (scripts/dev/*.py), browser automation via Playwright, and API interaction via FlowApiClient.
Sanitization: The skill mandates the use of extract_har_summary.py for redacting sensitive tokens before data is shared outside the local environment.
[COMMAND_EXECUTION]: The skill references local scripts (e.g., scripts/dev/spike_*.py) that drive browser automation via Playwright and the FlowApiClient. This is standard and expected functionality for a developer-centric debugging and measurement tool.

spike