ktl-docent
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local bash scripts (
knowledge-report.shandknowledge-feedback.sh) to process metadata and manage feedback. It explicitly warns against running scripts from the repository being analyzed, which effectively prevents remote code execution (RCE) attacks that could be triggered by untrusted repository content. - [DATA_EXPOSURE]: The skill implements strict guardrails to prevent the exfiltration of sensitive information. It instructs the agent to never include secrets, credentials, or connection strings in responses or feedback entries, recommending that only the file name and the type of value be mentioned.
- [INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of malicious instructions embedded in processed repository data. It includes a specific guardrail to treat all fetched content as data to be summarized or quoted, not as instructions.
- Ingestion points: Repository source files and the
.lokf/knowledge bundle files. - Boundary markers: Explicit instructions in
SKILL.mdto treat repository content strictly as text/summary. - Capability inventory: Execution of local bash scripts, reading repository files, and writing sanitized entries to
.lokf/feedback.md. - Sanitization: The
knowledge-feedback.shscript sanitizes input by stripping control characters and newlines, and collapsing whitespace before writing to the feedback file. - [EXTERNAL_DOWNLOADS]: The skill uses official CLI tools (
ghandglab) to fetch the user's login for attribution. These operations are performed using filters (jq,sed) to ensure only the username is retrieved and no sensitive profile data is exposed. These represent interactions with well-known, trusted services.
Audit Metadata