This is the template for the NFT Collector Copilot agent on Pinata. The OpenSea skill it depends on lives in ProjectOpenSea/opensea-skill and is published to ClawHub as opensea-marketplace. Skill changes do not belong here — open them against that repo. Template changes (workspace docs, manifest, README) belong here.
See README.md → Repository layout. The short version: manifest.json defines the agent, .openclaw/ configures the harness, workspace/ holds everything the agent reads at runtime.
- Branch from
main:git checkout -b your-change. - Edit. Most edits are to
workspace/*.mdormanifest.json. - Run validation locally before pushing — same checks CI runs:
jq . manifest.json > /dev/null jq . .openclaw/openclaw.json > /dev/null
- Open a PR. CI runs the same shape checks plus presence checks for required workspace files.
manifest.json has one version field — version (top-level integer) — which is the manifest schema version (currently 1). Don't touch it unless the schema itself changes.
The template doesn't carry its own version field; track template revisions through git tags instead:
git tag v1.0.1 && git push --tagsUse semver for tags:
- patch: doc tweaks, prompt fixes
- minor: new heartbeat steps, new memory schemas, new TOOLS.md fields
- major: breaking changes to memory layout, manifest shape, or env requirements
Match the existing voice: terse, numerate, defers to the skill rather than duplicating it. The agent reads these files every session — every line costs context. Cut anything that isn't load-bearing.
- Secrets,
.env*, real API keys, real wallet IDs. workspace/memory/*— runtime state, gitignored..claude/— local Claude Code settings.
Open an issue or ping the OpenSea team.