The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (scan.sh, quick-diff.sh, save-results.sh) to perform file system inventory, calculate version differences based on timestamps, and manage JSON result files.
[DATA_EXFILTRATION]: The skill accesses the local tool-use log (~/.claude/observations.jsonl) to calculate usage frequency for skills. While this file contains a history of agent activities, the data is used locally for auditing purposes and no network exfiltration was observed.
[PROMPT_INJECTION]: The auditing process utilizes a general-purpose subagent to evaluate the contents of other skills, which introduces a surface for indirect prompt injection. Maliciously crafted skill files could attempt to influence the auditor's judgment. Ingestion points: Skill content and frontmatter are read from the file system by scan.sh and passed to a subagent. Boundary markers: The subagent prompt uses [INVENTORY] and [CHECKLIST] markers to separate instructions from data. Capability inventory: The skill can propose file modifications, mergers, or deletions in Phase 4, requiring user confirmation before execution. Sanitization: There is no explicit sanitization of the content extracted from audited skills before it is processed by the AI subagent.

skill-stocktake