Ecosystem examples & OpenClaw- / ZeroClaw- / Hermes-oriented imports

Examples in examples/ecosystem/ are kept fresh via weekly auto-sync from upstream Clawflows & Agency-Agents repos (GitHub Actions: .github/workflows/sync-ecosystem.yml, Monday 04:00 UTC, plus manual workflow_dispatch). The job runs ainl import clawflows / ainl import agency-agents and opens a PR when conversions under examples/ecosystem/** change.

Recommended production stack: AINL graphs + AVM or general agent sandboxes

AINL provides deterministic graphs and declared capabilities; pair with AVM or other sandbox runtimes (microVM/container/agent-sandbox) for isolation. Integration remains additive and optional via execution_requirements, ainl generate-sandbox-config, and unified sandbox shim hooks.

Sync workflow & upstream ownership

We do not need admin access to Clawflows or Agency-Agents. The job only downloads public raw Markdown (same URLs as the CLI importer) and commits changes in this repository.

Some GitHub orgs block the default GITHUB_TOKEN from creating pull requests. If the sync job fails with “GitHub Actions is not permitted to create or approve pull requests”:

1. Create a fine-grained or classic PAT for a bot or maintainer account with contents: read/write and pull requests: read/write on this repo only (classic: repo scope for a private repo, or public-repo public_repo if applicable).
2. Add it as an Actions secret named GH_PAT (Settings → Secrets and variables → Actions).
3. Re-run the workflow — .github/workflows/sync-ecosystem.yml uses secrets.GH_PAT when set, otherwise github.token, for both peter-evans/create-pull-request and the gh label create step.

Never commit the PAT; rotate it if leaked.

What lives here

• Curated templates: examples/ecosystem/README.md — original.md, converted.ainl, per-folder README.md.
• CLI & flags: root README.md → Ecosystem & agent hosts (ainl import markdown, ainl import clawflows, ainl compile, --generate-soul, etc.). When parsing fails, the importer may emit a minimal_emit fallback stub so graphs still compile for review.
• MCP (Claude Code / Cursor / Gemini CLI / OpenClaw / ZeroClaw / Hermes): docs/INTEGRATION_STORY.md (Import Clawflows & Agency-Agents via MCP) and docs/operations/EXTERNAL_ORCHESTRATION_GUIDE.md §9 — tools ainl_list_ecosystem, ainl_import_clawflow, ainl_import_agency_agent, ainl_import_markdown. OpenClaw skill + bootstrap: docs/OPENCLAW_INTEGRATION.md · ZeroClaw skill: docs/ZEROCLAW_INTEGRATION.md · Hermes: docs/HERMES_INTEGRATION.md · Hermes Agent.
• Community PRs: .github/PULL_REQUEST_TEMPLATE/ (workflow / agent submission templates).

Converted graphs often include cron, sequential Call steps, and optional memory / queue hooks for OpenClaw-style bridges (sync workflow uses --no-openclaw-bridge for portable samples; re-import locally without that flag when you want bridge ops). OpenClaw users wire the same importer and MCP tools via skills/openclaw/ and ainl install-openclaw (~/.openclaw/openclaw.json, ainl-mcp); see OPENCLAW_INTEGRATION.md. ZeroClaw users consume the same surface via the ZeroClaw skill and ainl install-zeroclaw (~/.zeroclaw/mcp.json, ainl-mcp). Hermes Agent users compile to skill bundles and run via ainl_run: HERMES_INTEGRATION.md, skills/hermes/, ainl install-mcp --host hermes, upstream Hermes Agent.

See also

• OpenClaw skill & bootstrap: docs/OPENCLAW_INTEGRATION.md
• ZeroClaw skill & bootstrap: docs/ZEROCLAW_INTEGRATION.md
• Hermes Agent & bootstrap: docs/HERMES_INTEGRATION.md · Hermes Agent
• Benchmarks & viable leverage context: docs/benchmarks.md, BENCHMARK.md