ariffazil/AGI-emd-encode

AGI · emd-encode

The E in EMD. Bound to 333-AGI runtime. Doctrine: how to take raw input (observation / request / document / signal) and bind it into a typed, tri-witness-ready segment for downstream M and D.

When to invoke

• Before any reasoning chain that will produce a persistent artifact (skill draft, memory record, VAULT999 entry, agent-card mutation)
• Before handing work to another agent (arif_route, forge_execute_sealed, etc.)
• Before any LLM-generated output that the session will rely on

Encode contract

encode:
  - segment_id: ulid
    source_uri: "<originating input>"
    input_class: text | code | media | structured | signal
    raw_sha256: 12-char-prefix
    intent: "<what this segment should enable>"
    epistemic_label: OBS | DER | INT | SPEC
    witness_plan:
      human: required | opportunistic | none
      ai: required (always for AGI writes)
      external: required | opportunistic | none
    floor_projection: [F1, F2, F4, ...]   # which floors this will need to clear
    downstream_consumer: 333-AGI | 888-APEX | opencode | forge | well
    reversibility: reversible | irreversible | sovereign
    ttl_hours: number | null

Tri-witness readiness (F3)

An encoded segment is ready for the chain only when:

• 1 witness from each channel (Human / AI / External) is attached OR explicitly marked opportunistic
• Floor projection is asserted (the encoder's best guess for which F1-F13 will apply)
• Reversibility class is declared (reversible / irreversible / sovereign)

Substrate rule binding

Encode → Metabolize → Decode is the AAA substrate cycle. This skill is the ingress. Encoded segments feed AGI-emd-metabolize (memory promotion gates) and AGI-emd-decode (output decomposition + ground-truth binding).

Skipping encode = skipping witness readiness = Malu increment + 888_HOLD risk.

Not instead of

Distinct from AGI-emd-metabolize (forward = decision) and AGI-emd-decode (forward = bind). Encode is the first step, not the only step.

DITEMPA BUKAN DIBERI.