The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads local Claude Code session logs (typically found in ~/.claude/projects/) and user-provided matter.yml files. This access is necessary for its core functionality of clustering session events into billable time entries.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from session logs which could contain adversarial content.
Ingestion points: Session logs are ingested via the extractEventsWithHashes function in draft-entries.mjs.
Boundary markers: The generated reports use clear Markdown delimiters and structured HTML sections to separate metadata from log content.
Capability inventory: The skill is limited to local file system reads and writes and does not perform network operations or allow external content to influence system-level commands.
Sanitization: An esc function is used in renderAuditPacket to sanitize content for the HTML audit artifact, preventing cross-site scripting (XSS) in the generated packet.
[COMMAND_EXECUTION]: The included test suite (test/cli.test.mjs) uses node:child_process.spawnSync to execute the local CLI script. This is a standard practice for verifying command-line interface behavior during development and does not represent an arbitrary execution risk.
[EXTERNAL_DOWNLOADS]: The skill contains references to official ethics opinions from legal organizations (e.g., American Bar Association, State Bar of California) within the disclosures/ YAML files. These are informational URLs intended for user reference and do not trigger automated downloads or remote code execution.

billable-time-stephane-boghossian