dot-iu-cutter v0.4 — Cutter-Agent Design GPT Review

Date: 2026-05-16
Reviewer: GPT
Files reviewed: v0.4 8-file cutter-agent design package
Scope: Design review + open-decision ratification. No code or production action authorized by this review alone.

---

1. Verdict

v0_4_cutter_agent_design_status: PASS
agent_revision_needed: false
Tier_2_design_ready: true
code_authoring_allowed_after_explicit_prompt: true
production_CUT_VERIFY_allowed: false
credential_creation_allowed: false

The 8-file v0.4 cutter-agent design package is accepted. The recommended runtime spine is accepted: signal → MARK → REVIEW → CUT → VERIFY, with each phase as one atomic PG transaction and append-only ledger semantics.

---

2. Accepted Flow

flow:
  signal: external_trigger_or_backlog_signal
  MARK: decision_backlog_entry + history + dependency + sweep_log
  REVIEW: manifest_envelope + manifest_unit_block + review_decision
  CUT: cut_change_set + affected_row + executor dot_pair_signature DOT_991
  VERIFY: verify_result + verifier dot_pair_signature DOT_992
transaction_model: one_atomic_PG_transaction_per_phase
resumability: phase_independent_and_idempotent
rollback_model: forward_compensating_cut_change_set_not_physical_delete
canonical_address_alias: deferred_untouched_in_v0_4

---

3. Ratified Open Decisions

OD_1_idempotency_key_location:
  decision: payload_idempotency_key
  note: scenario_ref remains human/test label only, not the canonical idempotency key

OD_2_canonical_address_alias:
  decision: full_deferral_from_v0_4
  note: no alias rows, no alias writer, no alias resolver side-effect in v0.4

OD_3_writer_principal_count:
  decision: two_principals
  principals:
    - cutter_exec
    - cutter_verify
  note: real separation of duty at DB identity layer

OD_4_signal_source_contract:
  decision: separate_later_design
  note: v0.4 code authoring may expose a local/test input contract only; production signal source is out of scope

OD_5_retry_model:
  decision: append_only_chained_rows
  note: semantic retries create new rows chained by prior/superseded fields; no overwrite-as-retry

OD_6_cut_change_set_verifier_signature_id:
  decision: leave_NULL_in_CUT
  note: verifier signature belongs on verify_result; no VERIFY backfill into cut_change_set in v0.4

OD_SM_1_concurrency_guard:
  decision: compare_and_set_on_status
  note: no advisory lock in v0.4 unless dry-run proves CAS insufficient

OD_SM_2_S5_S7_inflight_states:
  decision: conceptual_transaction_internal_not_persisted
  note: persist only stable states and terminal/escalation outcomes

OD_SM_3_sweep_promotion_authority:
  decision: same_agent_logged_pass
  note: sweep authority stays within cutter agent but every sweep writes sweep_log + history

OD_CR_1_secret_custody_substrate:
  decision: VPS_env_pattern_for_future_credential_cycle
  note: no secret minted in design or code-authoring; dedicated secrets store deferred until scale/risk demands

---

4. Accepted Principal Strategy

writer_principals_future_cycle:
  cutter_exec:
    lane: MARK_REVIEW_CUT
    DOT: 991
  cutter_verify:
    lane: VERIFY
    DOT: 992
cutter_ro: read_only_never_write
forbidden:
  - workflow_admin_for_runtime_writes
  - directus_role_for_runtime_writes
  - single_writer_principal_for_exec_and_verify
  - credential_creation_inside_code_authoring

Credential creation remains a separate gated cycle. Code-authoring may define interfaces/config names, but must not mint credentials or edit .env.

---

5. Risk Decision

design_risk: STANDARD
runtime_risk: ELEVATED
high_risk_items_deferred:
  - canonicalization_alias_semantics
  - signing_scheme_cryptography
  - first_nonzero_write_to_governance_ledger
  - credential_execution

The next code-authoring cycle must remain dry-run/test oriented and must not perform production CUT/VERIFY.

---

6. Gate Effect

B_CODE_1_design_review_PASS: closed
B_CODE_2_open_decisions_resolved: closed
B_CODE_3_explicit_code_authoring_prompt: still_required
ready_for_code_authoring_prompt: true
ready_for_credential_cycle: false
ready_for_dry_run: false
ready_for_production_CUT_VERIFY: false

---

7. Required Next Package

The next package should be v0.4 code-authoring package, not production execution. It should create code artefacts only, for GPT review.

Required characteristics:

allowed:
  - repository code authoring
  - local/unit-test code where safe
  - no production database writes
  - no credential creation
  - no production CUT_VERIFY
  - no deploy
must_include:
  - CLI entrypoint design/implementation for dry-run only
  - DB adapter with transaction boundaries
  - idempotency helper
  - state transition validator
  - append-only ledger writer interfaces
  - canonicalization placeholder interface, no semantic expansion
  - signing placeholder interface, no real secret or production key
  - structured test harness plan

---

8. Status

v0_4_design_review: PASS
open_decisions: resolved
agent_self_advance: prohibited_without_explicit_prompt