@cleocode/agents

Part of CLEO. Start with the root README; run cleo <command> --help for any command.

Universal subagent protocol, canonical worker templates, and meta-agents for the CLEO ecosystem.

Scope (ADR-068 — v2026.5.30 and later)

Per ADR-068, this package ships exactly three surfaces:

1. cleo-subagent.cant — the universal protocol base every agent extends.
2. templates/ — five named worker templates with {{variable}} placeholders. Filename basename equals declared agent <name>: per the install-validator contract.
3. meta/ — meta-agents that synthesize other agents from project context.

CleoCode-team dogfood personas (the former cleo-prime, cleo-dev, cleo-historian, cleo-rust-lead, cleo-db-lead, cleoos-opus-orchestrator) live in .cleo/cant/agents/ in the cleocode repository and are NOT shipped to users.

Package Tree

packages/agents/
├── package.json
├── README.md                                 # this file
├── cleo-subagent.cant                        # universal protocol base
├── templates/
│   ├── project-orchestrator.cant             # coordinates the starter team
│   ├── project-dev-lead.cant                 # decides HOW, reviews workers
│   ├── project-code-worker.cant              # writes code within globs
│   ├── project-docs-worker.cant              # writes/edits documentation
│   └── project-security-worker.cant          # security review and audits
└── meta/
    ├── README.md
    ├── agent-architect.cant                  # meta-agent: synthesizes agents
    └── playbook-architect.cant               # meta-agent: synthesizes playbooks

The Universal Protocol Base: cleo-subagent.cant

Every CLEO agent extends cleo-subagent.cant. It defines:

• RFC 2119 constraints (BASE-001…BASE-007) — manifest append, no content in responses, cleo complete as the terminal, output-before-manifest ordering, focus before work, no fabrication, research linking.
• LOOM lifecycle — Spawn → Execute → Output → Return, with explicit stage-specific guidance injected at spawn time.
• Return format contract — three allowed response strings, everything else goes to files and the manifest.
• Error handling — status classification, retryable exit codes, staleness and evidence rules (ADR-051).

Canonical Worker Templates

The five worker templates are parameterized blueprints with {{variable}} placeholders. They MUST remain project-agnostic — no CLEO-internal references, no tool-chain assumptions beyond what the template explicitly parameterizes.

These five make a complete starter team: one orchestrator + one lead + three workers. For projects that need richer topologies, the agent-architect meta-agent (see below) synthesizes additional personas.

Naming Contract (ADR-068 Decision 1)

Every .cant filename basename MUST equal the agent <name>: declaration inside it. Templates use the project-<role> prefix to match the classifier output (packages/core/src/orchestration/classify.ts). This is enforced by the install validator at packages/core/src/store/agent-install.ts.

How Variables Work

CLEO uses mustache {{var}} syntax for template substitution, per ADR-055 D033.

Syntax

• {{name}} — simple variable
• {{object.key}} — dot-notation for nested values
• {{inputs.taskId}} — already used in starter .cantbook playbooks

Resolver Chain (ADR-068 Decision 5)

Variables resolve in priority order at spawn time (not install time):

1. Step bindings — highest priority; step-level bindings: shadow playbook bindings.
2. Playbook bindings — top-level bindings: field.
3. Session context — playbook_runs.bindings, task + epic identifiers, user.
4. Project context — .cleo/project-context.json, traversed via dot-notation.
5. Environment variables — CLEO_* or CANT_* prefix.
6. Default value — when SubstitutionOptions.defaultValue is set.
7. Missing — strict mode throws E_TEMPLATE_RESOLUTION.

Why Lazy (Spawn-Time)?

Templates install with {{...}} placeholders intact. Resolution happens inside orchestrateSpawnExecute right before composeSpawnPayload, which means:

• The same template can spawn differently under different bindings.
• BRAIN-provided context (mental-model slices, memory queries) can feed the resolver.
• Project context changes are picked up on the next spawn without reinstalling.

Auto-Installation on cleo init

Per ADR-068 Decision 3, plain cleo init (no flags) automatically walks @cleocode/agents/templates/ and calls installAgentFromCant() for each of the 5 worker templates. Each template is registered in signaldock.db.agents with tier='project'.

A fresh cleo init followed by cleo orchestrate spawn for any of the 5 worker roles succeeds without E_AGENT_NOT_FOUND.

The --install-seed-agents flag is preserved as a deprecated no-op alias with a deprecation notice.

Authoring a New Agent

1. Start from a template or from cleo-subagent.cant

cp packages/agents/templates/project-code-worker.cant \
   my-project/.cleo/cant/agents/my-worker.cant

Edit the agent name, tune the description + skills + tool list, replace {{variable}} placeholders with either literal values (if project-specific) or leave them for lazy resolution at spawn time.

2. Validate the CANT syntax

cleo cant validate my-project/.cleo/cant/agents/my-worker.cant

3. Install into the registry

cleo agent install my-project/.cleo/cant/agents/my-worker.cant

4. Verify resolver coverage

cleo agent doctor --json

Tier Precedence (Resolver Chain)

Agent resolution at spawn time walks tiers in order (ADR-055 / ADR-068):

1. project — {projectRoot}/.cleo/cant/agents/{agentId}.cant
2. global — ~/.local/share/cleo/cant/agents/{agentId}.cant
3. packaged — packages/agents/templates/{agentId}.cant
4. fallback — seed file on disk with no registry row
5. universal — cleo-subagent.cant synthesized envelope (ADR-068 Decision 6)

E_AGENT_NOT_FOUND is only thrown when cleo-subagent.cant itself is unreachable, indicating a corrupt installation.

Contract Guarantees

• Atomic install — packages/core/src/store/agent-install.ts wraps the .cant copy, agents row upsert, and agent_skills junction rewrite in a single BEGIN IMMEDIATE TRANSACTION.
• Idempotent seed install — packages/core/src/agents/seed-install.ts compares .seed-version against the bundled package.json version and returns early when they match.
• Doctor drift reporting — packages/core/src/store/agent-doctor.ts emits D-001…D-010 codes for orphan files, SHA mismatch, legacy paths, missing skills, and legacy JSON registries.

See Also

• ADR-055 — Agents Architecture + Meta-Agents (partially superseded by ADR-068)
• ADR-068 — Canonical Agent System (active canonical reference)
• Package boundary contract — canonical layering for every CLEO package

License

MIT — see LICENSE.