| name | skillwiki-ingest |
| description | Use this skill when the user wants to "ingest a document into SkillWiki", "add a skill from a file", "import experience into the skill wiki", "create a skill candidate from a script or trajectory", "batch import past skills", or "extract skills from docs". This skill guides the full ingest → audit → promote pipeline using the skillwiki CLI. |
| version | 0.1.0 |
SkillWiki Ingest
You help the user add raw experience, documents, scripts, or existing skill definitions into the SkillWiki knowledge base using the skillwiki CLI.
Prerequisites
The SkillWiki backend must be running. If it is not, start it first:
cd <repo-root>/skillwiki
venv/Scripts/activate
skillwiki serve --backend memory
Step 1 — Identify source type
Ask the user what they want to ingest if not already clear. Map it to one of these types:
| What the user has | SOURCE_TYPE |
|---|
| Operation logs, conversation traces, step-by-step records | trajectory |
| Tutorial, guide, specification document | document |
| API endpoint documentation | api_doc |
| Shell script or automation script | script |
| Existing skill JSON / JSONL files | past_skills |
Step 2 — Run ingest
skillwiki ingest run <SOURCE_TYPE> <file_path>
skillwiki ingest run <SOURCE_TYPE> "<text content>"
skillwiki ingest run <SOURCE_TYPE> <file_path> --create
skillwiki ingest run past_skills ./skills.jsonl --max-candidates 20 --create
The command prints each extracted candidate with its ID and name.
Step 3 — Audit the candidate
Run static checks (schema completeness, safety patterns, postcondition alignment):
skillwiki audit <candidate_id>
If audit fails, show the issues to the user and ask whether to fix the source and re-ingest, or proceed anyway.
Step 4 — Promote to released
Move the skill through the lifecycle:
skillwiki promote <candidate_id> S2
skillwiki promote <candidate_id> S3
skillwiki promote <candidate_id> S4
Or run the verify loop to auto-promote to S3:
skillwiki verify <candidate_id> --watch
Notes
- Ingest requires a live LLM connection — the pipeline calls the LLM to extract and normalize skills.
past_skills type accepts .json (single object or array) and .jsonl (one object per line).
- The
--create flag automatically creates S1 candidates; without it, candidates are only shown but not saved.