oma-recap
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses sensitive user information by reading conversation history files located at
~/.claude/history.jsonl. It also reads the agent's configuration file at.agents/oma-config.yaml. These actions are necessary for the skill's primary purpose of summarizing activity, and the data is stored locally in the.agents/results/recap/directory. - [COMMAND_EXECUTION]: The skill executes local system commands including
oma recap,jq, anddate. These are used to resolve time ranges, extract history records, and calculate timestamps for filtering logs. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from AI conversation logs which could contain adversarial instructions.
- Ingestion points: The skill reads from
~/.claude/history.jsonland the output of theoma recap --jsontool. - Boundary markers: There are no explicit delimiters or boundary markers used when interpolating historical prompt content into the reasoning context for theme analysis.
- Capability inventory: The skill has the ability to write files to the local filesystem (
.agents/results/recap/) and execute local CLI tools. - Sanitization: The skill performs length-capping (150 characters) and removes newlines from prompt content in its
jqfallback logic, but it does not filter for potential instruction-like text that could influence the agent's recap behavior.
Audit Metadata