ODF docs
GitHub ↗
Documentation menu

ODF Fast Lane Tasks

11 min read

This is the executable fast-lane task breakdown. Local policy integration is implemented where marked; representative Odoo validation and canary evidence remain pending.

#Work-unit rules

  • Keep each unit reviewable and independently verifiable.
  • Do not add a new canonical workflow or WorkType; use the existing small-change route for eligible features and preserve bugfix for fixes.
  • Fail closed on missing facts, risk signals, policy/evidence failures, or scope changes.
  • Do not run destructive database commands. All future Odoo commands must name the exact authorized -d <test_db>.

#1. Baseline instrumentation and measurement

#FL-01 — Establish the latency baseline

  • Depends on: none.
  • Status: Partial / in progress. The task-call baseline exists, but the end-to-end entry-to-final-gate baseline is unavailable.
  • Likely files: odf-plugin/odf-delegation-metrics.ts, plugins/odf-delegation.ts (only if current events cannot reconstruct entry-to-final-gate timing).
  • Intent: Use existing run/span JSONL telemetry to measure /odf-new and /odf-fix p50/p95, task-call count, phase duration, outcome, work type, model availability, validation, receipts, and escalation. Add only bounded allowlisted dimensions for entry level, micro-policy result, validation mode, and redundancy-scan disposition when required.
  • Verification commands/evidence: node "$PACK/scripts/odf-toolkit.js" metrics --days 14 --json; 2026-09-16 baseline report records the available task-call evidence and the end-to-end measurement gap. Confirm no raw prompt, path, secret, database value, or PII is emitted.
  • Out of scope: New telemetry storage, generic tracing framework, speculative cache, or behavior changes.

#FL-02 — Define the comparison cohort and rollback signal

  • Depends on: FL-01.
  • Status: Proposed / in progress. The cohort and rollback definition is documented, but it has not been executed or approved for canary use.
  • Likely files: docs/odf-fast-lane-plan.md, rollout configuration file identified by the implementation owner (no new framework).
  • Intent: Freeze the baseline window and define comparable cohorts from complete entry_started → verified_completed flows with matching work_type, Odoo version, workspace, and source-authority dimensions when available. Exclude lifecycle markers, spans, and joins from the denominator.
  • Proposed hold/rollback signals: >10% matched-cohort p95 regression; any protected/architectural risk incorrectly downgraded; any candidate/evidence/receipt binding failure; missing final-gate evidence; telemetry coverage below 95%; or an unexplained increase in errors, timeouts, or receipts. These thresholds are proposed and require explicit maintainer approval before canary.
  • Verification commands/evidence: Reviewable baseline table and cohort/rollback definition plus a dry-run dashboard/query definition; no code execution required for the documentation artifact.
  • Out of scope: Claiming an improvement before a canary or changing default rollout state.

#2. Entry-triage optimization

#FL-03 — Reuse known project facts before triage

  • Depends on: FL-01.
  • Likely files: command/odf-new.md, agent/odoo_orchestrator.md, odf-plugin/entry-triage.ts, plugins/odf-delegation.ts.
  • Intent: Pass the existing project module list, Odoo version, domain/module facts, test command, source roots, and approved Expectations into one deterministic triage call. Avoid repeating stable discovery during the same entry/continuation. Preserve the one grouped question for missing or unclear ICE facts.
  • Verification commands/evidence: Triage fixture matrix showing stable inputs produce the same result; evidence that unknown modules remain warnings and missing facts remain a question/standard result rather than an unsafe micro result.
  • Out of scope: Persistent general-purpose cache, changing the risk vocabulary, or letting the LLM choose work_type.

#FL-04 — Move triage ahead of expensive redundancy work

  • Depends on: FL-03.
  • Likely files: command/odf-new.md, agent/odoo_orchestrator.md.
  • Intent: Make the order explicit: health → known facts → triage → conditional redundancy scan. Defer or reduce redundancy scanning only for a clear, single-module, low-risk micro result; retain it for warnings, unknown modules, contradictory prior learning, ambiguity, and all elevated signals.
  • Verification commands/evidence: Orchestrator trace or structured event proving order; cases for clear micro, unclear, unknown module, contradictory learning, and protected-risk entries; no scan suppression for any protected signal.
  • Out of scope: Deleting the redundancy tool, globally disabling prior-learning checks, or deciding scope from a scan result without user disposition.

#FL-05 — Tighten deterministic micro eligibility

  • Depends on: FL-03.
  • Likely files: odf-plugin/entry-triage.ts, odf-plugin/odf-workflow.ts only if transition metadata needs a non-routing policy field, odf-plugin/entry-triage.test.ts.
  • Intent: Encode the bounded policy as data/validation, not a second route: exactly one known Odoo module, one functional domain, initial file forecast <=3, approved clear Expectations, no security/migration/payment/public-api/data-loss/PII or architectural signal. Keep protected signals monotonic and preserve existing canonical work types.
  • Verification commands/evidence: Pure classifier cases for each positive and negative condition, boundary at 3/4 files, missing facts, unknown module, explicit work type conflict, and every protected signal.
  • Out of scope: Broad semantic inference, automatic file-count promises, or adding micro-change to WORK_TYPES.

#3. Micro policy and route integration

#FL-06 — Define the inline BUILD contract

  • Depends on: FL-05.
  • Status: Implemented locally / live Odoo evidence pending. Contract locations: command/odf-new.md, command/odf-fix.md, and agent/odoo_orchestrator.md. No canary or rollout is enabled.
  • Likely files: command/odf-new.md, command/odf-fix.md, agent/odoo_orchestrator.md, plugins/odf-delegation.ts.
  • Intent: For eligible small-change, create one bounded BUILD prompt containing approved intent/Expectations, module/domain, file budget, targeted context, exact Odoo source roots, test command, source-authority requirements, evidence path, and executor/database guard. For /odf-fix, retain FIX/root-cause/regression evidence before applying the bounded BUILD policy.
  • Verification commands/evidence: Documentation evidence is the linked contract above; no runtime evidence, canary, rollout, or performance claim exists in this slice. Future captured-prompt verification must cover all required fields and must block or promote on missing source roots, database authorization, Expectations, test command, scope, authority, task, or validation evidence.
  • Out of scope: Direct generic task() calls, bypassing odf_delegate, or allowing implementation to choose its own policy.

#FL-07 — Integrate the policy without a parallel workflow

  • Depends on: FL-06.
  • Status: Implemented locally / live Odoo evidence pending. The existing small-change route selects the bounded executor only through a validated persisted policy; all other routes retain their canonical agents.
  • Likely files: odf-plugin/odf-workflow.ts, plugins/odf-delegation.ts, command/odf-new.md, agent/odoo_orchestrator.md.
  • Intent: Keep small-change as DECIDE -> BUILD -> VERIFY and bugfix as FIX -> BUILD -> VERIFY. Make the micro policy select an executor inside those routes; do not globally merge dependent phases. A micro host verifier still performs the canonical VERIFY transition and cannot skip BUILD/VERIFY ordering.
  • Verification commands/evidence: Route snapshots remain unchanged for all existing work types; odf_workflow_advance rejects missing/invalid validation; standard/ambiguous/elevated cases still launch full agent VERIFY; micro failures promote rather than silently pass.
  • Out of scope: New stages, new route matrices, cross-domain parallelization changes, or a global FIX→BUILD/BUILD→VERIFY coalescence.

#FL-08 — Preserve state, ledger, gate, receipt, and authority seals

  • Depends on: FL-07.
  • Status: Implemented locally / live Odoo evidence pending. Fast-lane delegation uses the existing policy, attempt, evidence, workflow, and receipt chokepoints.
  • Likely files: plugins/odf-delegation.ts, odf-plugin/odf-delegation-metrics.ts, command/odf-new.md, command/odf-fix.md.
  • Intent: Route every micro execution through the same selected artifact store, Policy Gate, fresh attempt ID, validation evidence, candidate binding, source authority, telemetry, and receipt code paths. Add no fast-path write that can create unbound state or settle a phase without evidence.
  • Verification commands/evidence: Unit/integration fixtures for duplicate attempt ID, running/completed phase, stale/mismatched evidence, pending receipt, source-authority failure, task timeout, empty result, and policy block; inspect persisted artifact refs and receipt refs.
  • Out of scope: Weakening policy/evidence schemas or replacing host validation with agent prose.

#4. Targeted Odoo validation

#FL-09 — Add a targeted-test command contract

  • Depends on: FL-06.
  • Status: Implemented locally / representative Odoo command evidence pending. Targeted evidence is phase- and kind-bound and cannot be reused as final full-gate evidence.
  • Likely files: agent/odoo_orchestrator.md, command/odf-new.md, command/odf-fix.md, plugins/odf-delegation.ts.
  • Intent: Use Odoo 19 --test-tags for module/class/method selection and --test-file for a specific Python file in the inner loop. Require exact -d <test_db>, authorized database context, exit code, output evidence, and the existing minimum command count by risk tier.
  • Verification commands/evidence: Future evidence records such as odoo-test with odoo-bin -d <test_db> --test-tags /<module> --stop-after-init or odoo-test with odoo-bin -d <test_db> --test-file <path> --stop-after-init; verify the host rejects missing -d, missing output, non-zero exit, stale evidence, or wrong candidate.
  • Out of scope: Destructive database setup, guessed databases, replacing the final module/CI suite, or changing Odoo test semantics.

#FL-10 — Keep the full final gate

  • Depends on: FL-09.
  • Status: Implemented locally / representative Odoo command evidence pending. Fast-lane VERIFY remains the full QA route and validates a distinct full evidence artifact.
  • Likely files: agent/odoo_orchestrator.md, plugins/odf-delegation.ts.
  • Intent: Require the configured full module/CI test command at the final gate after targeted inner-loop tests. Preserve the existing archive rule that cannot accept deferred, missing, or unrecorded module evidence.
  • Verification commands/evidence: A micro fixture proves targeted BUILD evidence is followed by the full configured command before successful archive; a missing final record blocks archive.
  • Out of scope: Treating targeted tests as sufficient final coverage or removing full agent VERIFY for standard/elevated work.

#5. Model and OpenCode steps policy

#FL-11 — Bound cheap classification and verification policy

  • Depends on: FL-05, FL-09.
  • Likely files: odf-registry.json, odf-plugin/odf-skills.ts, plugins/odf-delegation.ts, agent/odoo_orchestrator.md.
  • Intent: Consider a faster model and bounded OpenCode steps only for deterministic classification/host-verification prompts. Keep capable implementation routing for Odoo code, view authority, integrations, and any uncertain task. Make the policy explicit, opt-in/canary-controlled, and fail closed; do not silently override user-owned model/profile selection.
  • Verification commands/evidence: Profile-resolution output identifies model and step bound; fixtures prove no faster-model policy is selected for implementation when capability/risk requires the specialist; timeout/empty-result paths create receipts and escalate.
  • Out of scope: Global model replacement, unrestricted agent autonomy, silent step reduction for all phases, or model benchmarking without outcome metrics.

#FL-12 — Limit the micro prompt and context surface

  • Depends on: FL-06, FL-11.
  • Likely files: plugins/odf-delegation.ts, agent/odoo_orchestrator.md.
  • Intent: Keep inline prompts compact and stable: shared rules first, dynamic task facts late, only relevant module/source/test context, and a complete structured result contract. Do not spend the optimization budget on broad input trimming before reducing calls and generated output.
  • Verification commands/evidence: Compare prompt token estimate and task duration in metrics; verify required safety/source/evidence instructions remain present and task labels stay sanitized.
  • Out of scope: Prompt caching infrastructure, generic context-index rewrites, or removal of required source/evidence instructions.

#6. Tests and evidence

#FL-13 — Expand triage and route regression coverage

  • Depends on: FL-05, FL-07.
  • Status: Partial. Delegation safety and fast-lane route coverage are present; the broader triage/workflow/harness matrix remains to be completed.
  • Likely files: odf-plugin/entry-triage.test.ts, odf-plugin/odf-workflow.test.ts, scripts/odf-harness.test.ts.
  • Intent: Cover deterministic eligibility, protected-risk precedence, existing route snapshots, canonical stage ordering, and escalation behavior.
  • Verification commands/evidence: npx vitest run odf-plugin/entry-triage.test.ts odf-plugin/odf-workflow.test.ts scripts/odf-harness.test.ts; attach the test output and case matrix.
  • Out of scope: Removing existing route tests or testing only the happy path.

#FL-14 — Expand delegation safety and telemetry coverage

  • Depends on: FL-08, FL-09, FL-11.
  • Status: Partial. Fast-lane policy, candidate binding, evidence ordering, and delegation coverage are present; telemetry cohort validation remains pending.
  • Likely files: odf-plugin/odf-delegation.test.ts, odf-plugin/odf-delegation-metrics.test.ts if present/required, scripts/odf-harness.test.ts, plugins/odf-delegation.ts only for the implementation.
  • Intent: Prove policy/attempt/evidence/receipt/source-authority invariants remain active for micro, standard, and elevated routes. Verify p50/p95 inputs are bounded and sanitized.
  • Verification commands/evidence: npx vitest run odf-plugin/odf-delegation.test.ts scripts/odf-harness.test.ts; use fixture evidence with wrong candidate digest, wrong database, stale timestamp, missing output, non-zero exit, reused attempt, and pending receipt.
  • Out of scope: Live Odoo execution in the ODF harness unit suite or destructive database operations.

#FL-14H — Close local fast-lane hardening gaps

  • Status: Implemented locally; all listed local checks pass. Representative Odoo validation remains pending.
  • Likely files: plugins/odf-delegation.ts, odf-plugin/odf-delegation.test.ts, docs/odf-fast-lane-plan.md.
  • Intent: Require lowercase candidate-bound digests for every targeted evidence acceptance, and provide one explicit disable-fast-lane override that records its approver/reason in a fail-closed rollback marker without rewriting workflow state or artifacts.
  • Verification commands/evidence: Focused validation/override regressions, npm run typecheck, npm run test:unit, npm run test:yaml, registry validation, and git diff --check.
  • Out of scope: Odoo/PostgreSQL execution, canary activation, default-route changes, and odd/ bookkeeping.

#FL-15 — Validate against representative Odoo modules

  • Depends on: FL-10, FL-13, FL-14.
  • Likely files: No repository source change required unless a test fixture needs one; future evidence belongs in the review/CI system, not committed generated databases.
  • Intent: Exercise one clear model/view/test micro feature, one localized bugfix, one ambiguous request, and each protected-risk category. Include source-authority work and a case exceeding the file bound.
  • Verification commands/evidence: Authorized Odoo 19 targeted commands with -d <test_db>, followed by the configured full module/CI command; preserve .odf/validation-evidence-<change>.json and final verification report refs.
  • Out of scope: Production databases, destructive setup, broad performance claims from a single sample, or changing module behavior for the test.

#7. Documentation and rollout

#FL-16 — Update operational documentation after behavior is implemented

  • Depends on: FL-07, FL-09, FL-11, FL-14.
  • Status: Partial. Command and orchestrator contracts now match the local implementation; final rollout/metrics documentation review remains pending.
  • Likely files: command/odf-new.md, command/odf-fix.md, agent/odoo_orchestrator.md, docs/odf-fast-lane-plan.md.
  • Intent: Document the final policy name, exact eligibility, triage-before-redundancy order, targeted test templates, host-validation semantics, escalation triggers, kill switch, metrics query, and final-gate requirement. Keep all generated artifacts in English.
  • Verification commands/evidence: Markdown link check or repository documentation review; compare docs to actual route, plugin seals, and test evidence schema. No code path may rely on undocumented fast-lane behavior.
  • Out of scope: Claiming implementation in this planning change or duplicating the entire orchestrator contract.

#FL-17 — Run staged rollout and review metrics

  • Depends on: FL-01, FL-02, FL-15, FL-16.
  • Likely files: Existing rollout/configuration file only; no new framework or data store.
  • Intent: Execute measure-only → shadow → opt-in canary → controlled expansion. Review p50/p95, task count, targeted/full validation, receipt rate, timeout rate, escalation correctness, and artifact completeness at every stage.
  • Verification commands/evidence: node "$PACK/scripts/odf-toolkit.js" metrics --days 14 --json; attach before/after cohort report and explicit go/hold/rollback decision. Disable the policy when any acceptance threshold fails.
  • Out of scope: Automatic expansion, automatic retry around receipts, or silently changing the default route.

#Final acceptance checklist

  • [ ] ENTRY/TRIAGE was measured and optimized before generic plugin micro-optimization.
  • [ ] The existing small-change route remains canonical; no parallel full workflow or micro-change type was added.
  • [ ] Micro eligibility requires one known module, one domain, bounded files, clear approved Expectations, and no protected/architectural signal.
  • [ ] /odf-fix still requires diagnosis, root cause, and minimal regression before BUILD.
  • [ ] Health, preflight, Expectations, Policy Gate, attempt ledger, artifact persistence, source authority, validation evidence, receipts, and archive gates remain mandatory.
  • [ ] Triage runs before expensive redundancy scanning; clear low-risk micro entries alone may defer/reduce that scan.
  • [ ] Micro BUILD uses a bounded inline prompt and capable implementation routing where needed.
  • [ ] Inner-loop Odoo tests use --test-tags and/or --test-file with explicit -d <test_db>.
  • [ ] Full module/CI tests remain the final gate.
  • [ ] Standard, ambiguous, and elevated-risk work keeps full agent VERIFY.
  • [ ] FIX→BUILD and BUILD→VERIFY were not globally merged.
  • [ ] p50/p95 baseline and post-canary reports use comparable cohorts and include safety/error metrics.
  • [ ] Telemetry is bounded, sanitized, session-hashed, and free of secrets/PII.
  • [x] Rollback disables the micro policy through the audited disable-fast-lane operation without rewriting state, artifacts, evidence, or receipts.
  • [ ] npm run typecheck, focused Vitest tests, full repository tests, registry validation, and the representative Odoo evidence all pass in the implementation PR.
  1. perf(metrics): baseline fast-lane latency
  2. perf(triage): reuse facts before redundancy scan
  3. feat(policy): bound micro build and host validation
  4. test(odf): cover fast-lane safety invariants
  5. docs(odf): document fast-lane rollout

Each commit should remain independently reviewable. Do not commit generated databases, secrets, raw telemetry, or implementation claims with the documentation-only planning change.