codeburn-claude-cost-dashboard
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPERSISTENCEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Accesses sensitive interaction history stored in Claude Code session logs at
~/.claude/projects/. These logs contain full transcripts of agent activities, including source code, terminal outputs, and user prompts. - [PERSISTENCE]: Provides functionality to install a persistent background plugin for the macOS SwiftBar utility, which triggers periodic execution of the tool every 5 minutes.
- [EXTERNAL_DOWNLOADS]: Installs the
codeburnpackage via the npm registry and fetches model pricing data from the LiteLLM repository on GitHub. - [COMMAND_EXECUTION]: Executes shell commands via
npm,npx, andbrewto install and run the dashboard and its dependencies. - [INDIRECT_PROMPT_INJECTION]: Processes untrusted data from Claude session logs. Malicious instructions embedded in these logs (e.g., from tool outputs or previous prompt injections) could influence the tool's classification or reporting logic.
- Ingestion points: JSONL session transcripts in
~/.claude/projects/. - Boundary markers: None identified in the skill definition.
- Capability inventory: Local file system read access, network access for pricing updates, shell command execution, and persistent script installation.
- Sanitization: The documentation does not specify validation or sanitization of log contents before processing.
Audit Metadata