The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/live-copy-edit-agent.mjs is designed to execute the codex CLI tool using the --dangerously-bypass-approvals-and-sandbox flag. This pattern allows an AI sub-process to modify project source code with its internal safety and approval mechanisms disabled.
[COMMAND_EXECUTION]: The skill uses shell command execution to manage its environment and gather context:
scripts/pin.mjs creates and writes new skill configuration files (SKILL.md) into the agent's hidden harness directories (such as .claude, .cursor, and .github) to install command shortcuts.
Various scripts use execSync and spawn to run git commands and other node processes for project analysis.
Scripts like scripts/live-inject.mjs and scripts/live-wrap.mjs perform direct modifications to source code files and local environment configurations.
[PROMPT_INJECTION]: The live iteration feature in scripts/live-poll.mjs ingests steer events from a local helper server. These events contain a message field which the agent is explicitly instructed to interpret as design direction and execute, representing a surface for indirect prompt injection from the browser session.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to external sources:
scripts/context.mjs fetches version information from https://impeccable.style to perform update checks.
The detector engine in scripts/detector/engines/browser/detect-url.mjs uses Puppeteer to fetch and render content from arbitrary external URLs for analysis purposes.

impeccable