The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/runtime_bootstrap.py file uses os.execve() to replace the current process with a new Python process from a specific virtual environment located in the user's home directory (~/.claude/.fetching-salesforce-docs-runtime). This is a legitimate environment isolation technique used to ensure dependencies like Playwright are available without manual user configuration.
[EXTERNAL_DOWNLOADS]: The skill relies on playwright and playwright-stealth Python packages. These tools are used to render and extract content from documentation sites. During setup, Playwright typically downloads required browser binaries (Chromium) to the local machine.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests content from public web pages (ingestion points: Salesforce documentation URLs). The retrieval playbook provides boundary markers by instructing the agent to verify that the extracted content matches exact requested identifiers and to reject landing pages or shell results. The capability inventory includes local runtime management and network fetching. Sanitization is performed within the extraction scripts through text normalization and the removal of navigation noise before the data is presented to the agent.
[DATA_EXFILTRATION]: Network activity is restricted to fetching documentation from official Salesforce domains such as developer.salesforce.com and help.salesforce.com. The scripts return the extracted content locally to the agent and do not exhibit patterns of transmitting sensitive local data to unauthorized external servers.

platform-docs-get