task-track

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes several external programs to perform its tracking and workspace functions.
  • It spawns git via child_process.spawnSync in lib/shared-snapshot.cjs to read objects and repository state.
  • It spawns browser processes (Google Chrome or the system default) in lib/browser-launch.cjs to open the local workspace.
  • It spawns terminal applications (such as Terminal.app, cmd.exe, or various Linux terminal emulators) in lib/terminal-launch.cjs to run the server component.
  • [EXTERNAL_DOWNLOADS]: The skill manages its own environment by performing automated package installations.
  • In lib/package-setup.cjs, it executes npm ci with flags --omit=dev, --ignore-scripts, and --no-audit to install the yaml dependency if missing.
  • [DYNAMIC_EXECUTION]: The skill dynamically generates and executes shell scripts to facilitate the local workspace launch.
  • In lib/terminal-launch.cjs, it constructs .sh, .cmd, or .command scripts based on the user's operating system, writes them to tmp/task-tracking/launch/, and executes them.
  • These generation routines include safeguards such as POSIX and Windows-specific character filtering to prevent command injection into the generated scripts.
  • [INDIRECT_PROMPT_INJECTION]: The skill serves as a surface for indirect prompt injection because it ingests and processes untrusted project data.
  • Ingestion points: scripts/task-track.cjs (via readItem, readProgress) and lib/shared-snapshot.cjs read Markdown work records directly from the project directory.
  • Boundary markers: No explicit delimiter or "ignore instructions" tags are used when interpolating record content into the agent's context, although the skill instructions advise the agent to treat record text as data.
  • Capability inventory: The skill has capabilities for file system writes (executeOperation), starting a local web server (startWorkspace), and spawning external processes (launchBrowser, launchTerminal).
  • Sanitization: The skill utilizes a sanitized() helper function to filter tool outputs before returning them to the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 10:12 AM