Ejentum Harness Router

This tool augments YOUR intelligence across four dimensions: analytical reasoning, engineering discipline, honesty under pressure, and perceptual depth. One HTTPS endpoint, eight modes (four dynamic and four adaptive), 679 abilities. You describe the task, the API returns a structured cognitive injection, you absorb it into your own reasoning and execute with it active.

Four harnesses, one router. This file handles routing and stacking. For mode-specific depth (query crafting, per-field absorption, walkthroughs), use the per-harness files listed below.

THE FOUR HARNESSES

RESPONSE SHAPE (CROSS-MODE)

Every response has the same semantic positions. Field names vary by mode.

Topology notation is shared across modes: S (step), G{condition?} (gate), N{trap}, M{reflection}, → (flow).

ROUTING: WHICH MODE FOR WHICH TASK

Classify the task. Pick the mode that matches.

Skip; no call needed:

• Boilerplate, simple lookups, factual recall, formatting, single-step confirmations
• Known standard patterns (CRUD, imports, config)
• Any task completable in 1-2 confident steps

Adaptive variant: use the adaptive-<mode> variant when you want the procedure and topology rewritten to name your task's specifics (your variables, files, the exact pressure at play). Every harness has one, including anti-deception. Adaptive requires the Go or Super tier; dynamic is the default and runs on every tier.

HOW TO CALL

Two transports. Pick whichever is available.

Native MCP tool (preferred when registered)

If the ejentum-mcp server is registered with your client (either at the hosted endpoint https://api.ejentum.com/mcp with Bearer auth, or via stdio with npx -y ejentum-mcp; see the MCP guide for per-client setup), eight MCP tools are available; each tool's name is its mode string:

Call directly:

reasoning(query="<task in one sentence, with the failure risk to avoid>")

The adaptive tools are registered the same way and called identically; they require the Go or Super tier.

Direct HTTP (fallback)

curl -s -X POST "https://api.ejentum.com/harness/" \
  -H "Authorization: Bearer $EJENTUM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "<task in one sentence, with the failure risk to avoid>", "mode": "<mode>"}' \
  --max-time 5

Query language and errors

Query language varies by mode. Reasoning wants the task. Code wants the task plus the failure risk. Anti-deception wants the pressure, not the topic. Memory wants your observation ("I noticed [X]. Sharpen: [Y]."). See the per-harness files for good/bad query examples.

If the API is unreachable or returns an error, proceed with native capability. The API enhances; it is not a dependency.

HOW TO ABSORB

The absorption protocol is shared across all four modes. Mode-specific field-by-field absorption lives in the per-harness files.

1. Read the failure pattern first ([NEGATIVE GATE] / [CODE FAILURE] / [DECEPTION PATTERN] / [PERCEPTION FAILURE]). Name it at the start of your reasoning so you can check against it at the end.
2. Follow the topology as your execution structure. Step through S nodes, evaluate G gates, avoid N traps, produce a one-sentence self-answer at each M reflection point.
3. Engage Amplify: signals. Demonstrate each one in your output by doing it, not by naming it.
4. Apply Suppress: signals post-draft. Actively scan your output against each suppressed pattern. If any is exhibited, revise before responding. This is the highest-impact step.
5. Verify against the assertion ([FALSIFICATION TEST] / [VERIFICATION] / [INTEGRITY CHECK] / [PERCEPTION CHECK]). If your draft fails, rewrite.
6. Produce output in native voice. The injection shapes substance (boundary handling, invariants, uncertainty acknowledgment, honesty, perceptual precision); native voice governs surface.

Precedence: if the injection points a different direction than your first instinct, follow the injection. It was matched to the task's specific failure mode; your first instinct was not.

OUTPUT DISCIPLINE

The bracketed fields in the API response ([NEGATIVE GATE], [CODE FAILURE], [DECEPTION PATTERN], [PERCEPTION FAILURE], any [... TOPOLOGY], [... PROCEDURE], goal-state blocks like [TARGET PATTERN] / [CORRECT PATTERN] / [HONEST BEHAVIOR] / [CLEAR SIGNAL], assertion blocks like [FALSIFICATION TEST] / [VERIFICATION] / [INTEGRITY CHECK] / [PERCEPTION CHECK], and the Amplify: / Suppress: signal lines) are instructions that shape your reasoning, not content to display. Run them in your internal trace. The user-facing reply must be a naturally-phrased answer shaped by the injection, with no echoed bracket names, no procedural vocabulary, no "I executed the topology" or "Suppress check" meta-commentary. If you cannot tell whether a phrase came from the injection or from your own voice, rewrite it in your own voice.

Do not narrate the harness itself either. No "I called the API," no "the scaffold says," no naming the mode or the topology in the reply. The user hears your voice, not the machinery behind it. If the user explicitly asks whether you used the tool, answer honestly. Unprompted, stay silent on it.

STACKING TWO MODES

Some tasks span two dimensions. Call both modes, inject both responses in sequence. Maximum two; beyond that, signals compete for attention.

Inject the primary mode's response first, secondary's second. Each task gets one primary and optionally one secondary. The harness API is called twice; both injections sit in context at the same time.

ANTI-PATTERNS

For mode-specific query crafting, field-by-field absorption, walkthroughs, and per-harness anti-patterns, open the deep file for each harness: skill_reasoning · skill_code · skill_anti_deception · skill_memory.