agentleFS
hushh-labshushh-labs/hushh-research/.agents/skills/backend-agents-operons/SKILL.md

backend-agents-operons

Skill40 starsChanged 4 days agoApache-2.0View on GitHub

Use when changing backend agents, operons, ADK surfaces, or orchestration contracts inside the consent-protocol runtime.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Avoid overlap with backend-runtime-governance and security-audit.

    line 12
  2. Distinctive:

    Keep orchestration docs and relevant tests aligned with the implementation.

    line 56
  3. Very specific:

    If the request is still broad or ambiguous, route it back to backend.

    line 75
  4. Very specific:

    If the task becomes service/runtime boundary work, use backend-runtime-governance.

    line 76
  5. Very specific:

    If the task becomes trust or consent enforcement work, use security-audit.

    line 77

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Hussh Backend Agents Operons Skill

Purpose and Trigger

  • Primary scope: backend-agents-operons
  • Trigger on agents, operons, ADK surfaces, orchestration boundaries, or tool/agent runtime changes.
  • Avoid overlap with backend-runtime-governance and security-audit.

Coverage and Ownership

  • Role: spoke
  • Owner family: backend

Owned repo surfaces:

  1. consent-protocol/hushh_mcp/agents
  2. consent-protocol/hushh_mcp/operons
  3. consent-protocol/hushh_mcp/tools
  4. consent-protocol/hushh_mcp/hushh_adk
  5. consent-protocol/hushh_mcp/adk_bridge
  6. consent-protocol/hushh_mcp/one_adk

Non-owned surfaces:

  1. backend
  2. security-audit
  3. repo-operations

Do Use

  1. Agent and operon ownership or orchestration changes.
  2. ADK integration boundaries and backend tool surfaces.
  3. Agent-runtime docs and tests tied to orchestration behavior.

Do Not Use

  1. Broad backend intake where the correct spoke is still unclear.
  2. Generic service-layer placement or API wire contract changes.
  3. Trust-policy work where authorization or vault boundary is the primary concern.

Read First

  1. consent-protocol/docs/reference/agent-development.md
  2. docs/architecture/one/one-agent-hierarchy.md
  3. consent-protocol/docs/reference/kai-agents.md
  4. docs/architecture/ai/README.md

Workflow

  1. Confirm whether the change is about orchestration, tool surface, or agent packaging before editing code.
  2. Keep orchestration docs and relevant tests aligned with the implementation.
  3. Route trust, consent, or scope-enforcement questions into security-audit.
  4. For pod support, follow agent-development.md's "One authored fleet, explicit runtime dependencies" procedure: reuse the agent and inject dependencies; verify authority and fail-closed behavior through the existing pod ledger. Follow the dev pod runbook's startup qualification for provider-free rehearsal, incarnation recheck, the owned control socket and unchanged timeout deadlines.
  5. Separate portable shared-runtime corrections from pod wiring before any ADK worktree transfer. Registration or local synthetic tests alone do not establish pod completion. Private GCP placements share one image and PrivateGcpDeployment; follow deployment-standard.md's hosting-authority contract. Hosted custody is not user-owned cloud; its explicit dev inference binding neither widens the BYOC bridge nor qualifies admission.
  6. For Computer Use, read consent-protocol/docs/reference/private-browser-runtime.md. One and Browser share the owner container. Qualify the exact host/image and trusted Chromium mode (GCP enabled; Azure Container Apps explicitly disabled), native model transport, owner authority, lifecycle and ephemeral-screen handling before activation. Do not claim Browser-to-One containment, silently change execution mode or create a second approval ledger.
  7. Browser PKM selection requires independent model-processing and website-disclosure consent. Use the existing scoped-export and action-ledger ports; keep remembered sign-in state outside ADK context. The browser runtime reference owns manual login, Forget and recovery restrictions.
  8. Scheduled pod work keeps the sealed recovery log and existing effect owner authoritative. Native task checkpoints and pending calls stay in that same sealed journal, never chat-key/shared session storage. Construct native task mode with its owned output schema; validate the matching finish call/response, and never let a model summary settle a pending or uncertain effect. Freeze the sealed AI choice and destination for restart; process-local cache counters are not durable identity. Retain original invocation and confirmation markers; exact ledger approval remains mandatory and uncertain dispatch cannot resume. Read the Scheduled tasks section of docs/operations/hosting/private-files-library.md; bind approved revisions, current owner/incarnation and authorization expiry. Identifier-only due-time delivery is not model inference. Refuse unqualified cloud timers and uncertain-effect replay.
  9. Native pod Google grants require exact project/account lifecycle handling: confirm project-wide revocation before fresh authorization, then remove fenced legacy hub credentials without revoking the new grant. Connector credentials, tool reviews and notification work keep their existing owners; metadata feeds cannot carry private observations. Read the dev pod runbook for notification checkpoint and native-client prerequisites.
  10. Private command and MCP recovery must retain original expiry, exact owner/pod authority and dispatch-before-effect journals. Read the streaming implementation guide before changing those adapters; native resume must not replay a user turn or revive an uncertain effect.

Files retains one authored specialist across placements. Review the Files storage/processing boundary before changing its tool ports: Shared receives a selected foreground projection; pod execution keeps its own admitted storage and background authority. Generated registries never substitute for either boundary's live acceptance.

Handoff Rules

  1. If the request is still broad or ambiguous, route it back to backend.
  2. If the task becomes service/runtime boundary work, use backend-runtime-governance.
  3. If the task becomes trust or consent enforcement work, use security-audit.

Required Checks

cd consent-protocol && uv run pytest tests/agents -q
cd consent-protocol && uv run pytest tests/test_hushh_adk_foundation.py -q
cd consent-protocol && uv run python scripts/verify_agent_hierarchy_contract.py

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.