| name | mma-rag |
| description | Operate the locally running Tessmora knowledge service through its bundled deterministic mma-rag CLI. Use when Codex needs to list or create Tessmora knowledge bases, inspect indexed files, upload and index local documents/images/audio/video, wait for ingestion, retrieve compact multimodal evidence, or answer questions grounded in Tessmora. |
Tessmora (mma-rag CLI)
Use the bundled scripts/mma-rag CLI. The command and environment-variable prefix retain the mma-rag name for compatibility after the product rename to Tessmora. Resolve the script relative to this SKILL.md, invoke it by absolute path, and expect JSON on stdout for every command.
Workflow
- Run
scripts/mma-rag health before the first operation.
- If the service is unavailable, report that Tessmora must be started. Do not start or stop it unless the user asks.
- Run
kb list when the target knowledge base ID is unknown.
- Use
ingest files with absolute paths to add local multimodal files.
- Save every returned
processing_id and run ingest wait or ingest status.
- Treat content as searchable only after every relevant job reports
completed.
- Use
search when the user wants evidence or when Codex will synthesize the result.
- Use
ask when the user explicitly wants Tessmora to generate the grounded answer.
- Preserve returned file names, page numbers, and audio/video time ranges in the final response.
Safety
- Create knowledge bases or upload files only when the user's request authorizes the mutation.
- Never delete or overwrite Tessmora content; this skill exposes no deletion command.
- Upload only user-designated files. The CLI rejects files outside configured safe roots and common credential paths.
- Do not claim ingestion succeeded when a job is queued, processing, failed, missing, or timed out.
- Do not expose raw CLI diagnostics or secrets in the final response.
Command details
Read references/cli-reference.md when selecting flags, handling errors, filtering by files/modalities, or configuring safe upload roots.
