- name
- openclaw-success-skill-publisher
- description
- Capture successful OpenClaw interactions and convert them into reusable skills with an optimized execution path summary, then publish to ClawHub and distribute bilingual sharing posts to Moltbook, Zhihu, Xiaohongshu, or other configured channels. Use when the user asks to productize a completed OpenClaw workflow into a reusable skill and push it to distribution platforms.
- version
- 1.0.3
- metadata
- openclaw
- requires
- env
- bins
- primaryEnv
- CLAWHUB_API_TOKEN
OpenClaw Success Skill Publisher
Convert a successful user-agent interaction into a production-ready skill and distribute it.
Security And Runtime
- Require explicit user approval before any real external publish.
- Treat missing publish credentials as
draft-onlymode. - Never leak secrets in generated summaries.
- Use local output directory as source of truth for every publish payload.
Required Inputs
- A success record JSON containing:
- title - user_goal - steps[] - outcome.completed
- Optional:
- deliverables[] - context - outcome.metrics - language
Environment Variables
CLAWHUB_API_BASE(example:https://api.clawhub.ai)CLAWHUB_API_TOKENMOLTBOOK_WEBHOOK_URLZHIHU_WEBHOOK_URLXIAOHONGSHU_WEBHOOK_URL
In --dry-run, unset publish/share targets are skipped and local drafts are still generated. In real publish mode (without --dry-run), missing required env for a selected target should fail fast.
Workflow
- Build distilled knowledge from a success case.
python3 scripts/pipeline.py \
--input examples/success_case.json \
--output outputs/run-001 \
--dry-run- Review generated artifacts:
summary.md: what happened and why it workedoptimal_path.md: shortest high-confidence implementation pathgenerated_skill/SKILL.md: reusable skill definitiongenerated_skill/agents/openai.yaml: skill card metadatashare_payloads/*.md: platform-tailored bilingual posts
- Publish after explicit approval.
python3 scripts/pipeline.py \
--input examples/success_case.json \
--output outputs/run-001 \
--publish-clawhub \
--share moltbook zhihu xiaohongshuRules
- If
outcome.completedis false, stop and request more evidence. - Prefer deterministic step extraction over freeform storytelling.
- Rank step importance by observed success signal + dependency position + efficiency.
- Generate both Chinese and English sharing copy.
- In
--dry-run, if publish/share endpoint is missing, persist payloads locally and continue. - In real publish mode, require env for selected targets and fail fast when missing.
- Preserve reproducibility with placeholder env names and command templates only. Never embed real secret values in generated artifacts.
Output Contract
Always produce:
summary.mdoptimal_path.mdgenerated_skill/share_payloads/publish_report.json