claude-tracker-suite
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONOBFUSCATIONEXTERNAL_DOWNLOADSPERSISTENCECOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill indexes and processes conversation transcripts which contain untrusted content from previous user and assistant turns.
- Ingestion points:
scripts/index-transcripts.jsandscripts/search-sessions.jsread NDJSON transcripts from the~/.claude/projects/directory. - Boundary markers: The skill does not implement explicit delimiters or 'ignore' instructions when presenting search results or resuming sessions.
- Capability inventory: The suite has the ability to execute shell commands, resume sessions, and open new terminal tabs/splits with automated input.
- Sanitization: Content is filtered for 'transcript noise' (e.g., system reminders), but user-provided message text is stored and searchable without extensive sanitization.
- [OBFUSCATION]: Unicode Line Separator characters (U+2028) are used in several test files and fixtures (
tests/read-lines.test.js,tests/extract.test.js,tests/fixtures/rich.jsonl). Analysis confirms these are present as data fixtures to verify the robustness of the transcript parsing logic under various encoding conditions. - [EXTERNAL_DOWNLOADS]: The
scripts/backfill-summaries.jsscript makes network requests to the OpenRouter API (openrouter.ai) to generate session summaries. This feature is disabled by default and requires explicit user activation and configuration of API keys. - [PERSISTENCE]: The suite includes multiple
launchdagents (.plistfiles) designed to perform periodic background tasks, such as workspace snapshots, transcript indexing, and database maintenance. These are documented for manual user installation inreferences/daemon-setup.md. - [COMMAND_EXECUTION]: Several scripts utilize system automation tools on macOS (
osascript,pbcopy,pbpaste) to manage terminal state and simulate user input for tab creation and split-pane management.
Audit Metadata