Ouroboros Evolve Loop

ooo evolve - Evolutionary Loop

Description

Start, monitor, or rewind an evolutionary development loop. The loop iteratively refines the ontology and acceptance criteria across generations until convergence.

Flow

Gen 1: Seed(O₁) → Execute(all nodes) → Judge → Gate
Gen 2: Wonder(failed nodes) → Reflect(active nodes only) → Execute(active) → Gate(all)
Gen 3: Repeat with a smaller active set; frozen PASS nodes are reverified, not regenerated
...until the outcome gate passes, the working set stagnates, ontology converges,
or max 30 generations is reached

Usage

Start a new evolutionary loop

ooo evolve "build a task management CLI"

Fast mode (ontology-only, no execution)

ooo evolve "build a task management CLI" --no-execute

Check lineage status

ooo evolve --status <lineage_id>

Record an isolated full-graph benchmark control

Call ouroboros_evolve_step for an existing Gen 2+ lineage with benchmark_control: true, execute: true, and an explicit clean Git project_dir. Use a distinct clean worktree at the treatment commit. Normal evolve calls keep benchmark_control false and never launch a control arm.

Rewind to a previous generation

ooo evolve --rewind <lineage_id> <generation_number>

Instructions

Load MCP Tools (Required before Path A/B decision)

The Ouroboros MCP tools are often registered as deferred tools that must be explicitly loaded before use. You MUST perform this step before deciding between Path A and Path B.

1. Use the active runtime's tool-discovery capability to find and load the evolve MCP tools:

   tool discovery query: "+ouroboros evolve"

2. The tools will typically be named with prefix mcp__plugin_ouroboros_ouroboros__ (e.g., ouroboros_evolve_step, ouroboros_interview, ouroboros_generate_seed). After runtime tool discovery returns, the tools become callable.
3. If the tools are callable — already exposed, or loaded by discovery — proceed to Path A. An empty discovery result for already-exposed tools is expected, not a failure. Proceed to Path B only if they are genuinely absent (no Ouroboros MCP server).

IMPORTANT: Do NOT skip this step. Do NOT assume MCP tools are unavailable just because they don't appear in your immediate tool list. They are almost always available as deferred tools that need to be loaded first.

CRITICAL — deferred-schema guard (prevents "Invalid tool parameters"): This skill makes ouroboros_* MCP calls across multiple turns, and each turn runs in a fresh tool context. A deferred tool's schema loaded on one turn is NOT guaranteed to still be loaded on the next. If you call any ouroboros_* MCP tool while its schema is not loaded in the current turn, the runtime rejects the call with "Invalid tool parameters" before it ever reaches the server. Therefore: immediately before EVERY ouroboros_* MCP call in this skill, re-run the tool-discovery load query for the specific MCP tool or documented tool family you are about to call. Use "+ouroboros evolve" for ouroboros_evolve_step, ouroboros_lineage_status, and the evolve flow's documented tool family; use "+ouroboros interview" before ouroboros_interview, "+ouroboros seed" before ouroboros_generate_seed, and "+ouroboros lateral" before ouroboros_lateral_think. If a load returns no matching tool (and the tool is not already callable — an empty load for an already-exposed tool is an expected no-op, not absence), switch to the documented fallback / Path B instead of retrying the failing call.

Path A: MCP Available (loaded via runtime tool discovery above)

Starting a new evolutionary loop:

1. Parse the user's input as initial_context
2. Run the interview: call ouroboros_interview with initial_context
3. Complete the interview (3+ rounds until ambiguity ≤ 0.2)
4. Generate seed: call ouroboros_generate_seed with the session_id
5. Call ouroboros_evolve_step with:
   • lineage_id: new unique ID (e.g., lin_<seed_id>)
   • seed_content: the generated seed YAML
   • execute: true (default) for full Execute→Evaluate pipeline, false for fast ontology-only evolution (no seed execution)
   • benchmark_control: false (default). Set true only for a deliberate Gen 2+ full-graph control in an explicit clean Git project/worktree.
6. Check the action in the response:
   • continue → Inspect active_ac_indices, then call ouroboros_evolve_step again with just lineage_id. Only active failed/reopened nodes evolve; frozen PASS nodes stay immutable and are boundary-reverified.
   • ontology_stable → This is not success. Call ouroboros_evolve_step again for the same lineage_id with execute: true so the stable Seed goes through Execute→Evaluate. Do not call standalone evaluate or report convergence before that step returns converged.
   • converged → Evolution complete! Display final ontology
   • stagnated → Ontology unchanged for 3+ gens. Consider ouroboros_lateral_think
   • exhausted → Max 30 generations reached. Display best result
   • failed → Check error, possibly retry. If the error reports an expired lineage owner, first confirm that the prior owner process is dead, then make one explicit recovery call with recover_expired_claim: true. Never set this flag for a merely slow or still-running owner.
7. Repeat step 6 while action is continue. Treat ontology_stable as the explicit transition above, then process the execute: true response using step 6. Stop only on converged, stagnated, exhausted, or failed.
8. When the loop terminates, display a result summary with next step:
   • converged: ◆ Current state → next: Ontology converged! Run ooo evaluate for formal verification
   • ontology_stable: ◆ Current state → next: Run the same lineage with execute=true for Execute→Evaluate; this is not verified convergence yet
   • stagnated: ◆ Current state → next: ooo unstuck to break through, then ooo evolve --status <lineage_id> to resume
   • exhausted: ◆ Current state → next: ooo evaluate to check best result — or ooo unstuck to try a new approach
   • failed: ◆ Current state → next: Check the error above. ooo status to inspect session, or ooo unstuck if blocked

Checking status:

1. Call ouroboros_lineage_status with the lineage_id
2. Display: generation count, ontology evolution, convergence progress

Rewinding:

1. Call ouroboros_evolve_step with:
   • lineage_id: the lineage to continue from a rewind point
   • seed_content: the seed YAML from the target generation (Future: dedicated ouroboros_evolve_rewind tool)

Path B: Plugin-only (no MCP tools available)

If MCP tools are not available, explain the evolutionary loop concept and suggest installing the Ouroboros MCP server. See Getting Started for install options, then run:

ouroboros mcp serve --runtime claude-cli

Then add to your runtime's MCP configuration (e.g., ~/.claude/mcp.json for Claude Code).