roadmap-work-items-2026-06-04.md

Roadmap Work Items - 2026-06-04

Purpose

This work log turns the current strategy into concrete execution chunks.

Work Items

TASK-001: Close TUI / CLI semantic drift

• Goal: Make the same chat, cost, undo, root, and resume commands behave consistently across surfaces.
• Scope: TUI chat path, controller-backed shared state, help text, and surface-specific fallback wording.
• Dependencies: controller semantics, current TUI state model.
• Acceptance Criteria: same user command produces same trust semantics on CLI and TUI; fallback paths are explicitly labeled.
• Tests: regression coverage for cost accumulation, undo scope, and root precedence.
• Files likely touched: teaagent/tui/__init__.py, teaagent/chat_session_controller.py, surface docs.
• Risk: high
• Parallelizable: no
• Human Review Required: yes

TASK-002: Make undo wording and behavior honest

• Goal: Separate journal-based undo from checkpoint restore in both code and docs.
• Scope: command wording, current-status docs, recovery guide, and TUI help text.
• Dependencies: TUI controller alignment.
• Acceptance Criteria: user can tell which undo path is used before running it.
• Tests: undo regression tests on both live and fallback paths.
• Files likely touched: teaagent/tui/__init__.py, docs/recovery-and-continuity-guide.md, docs/daily-driver-current-status.md.
• Risk: high
• Parallelizable: yes
• Human Review Required: yes

TASK-003: Fix cost truth on the daily surfaces

• Goal: Ensure /cost, budget bars, and run summaries never show a fake zero or stale local state.
• Scope: shared cost source, display formatting, and cost documentation.
• Dependencies: shared controller state.
• Acceptance Criteria: cost reflects actual run spend after a real task execution.
• Tests: live task accumulation regression, formatting regression, budget display check.
• Files likely touched: teaagent/chat_session_controller.py, teaagent/tui/__init__.py, tests/test_tui.py.
• Risk: high
• Parallelizable: no
• Human Review Required: yes

TASK-004: Strengthen first-hour onboarding

• Goal: Make the first successful use path obvious without deep architecture reading.
• Scope: current-status front door, quick-start docs, and recovery pointers.
• Dependencies: stable trust-path docs.
• Acceptance Criteria: a new user can discover the safe first run, where to look for current status, and what to do on failure.
• Tests: docs consistency and acceptance references.
• Files likely touched: docs/daily-driver-current-status.md, docs/tui-daily-driver-guide.md, docs/use-cases.md.
• Risk: medium
• Parallelizable: yes
• Human Review Required: no

TASK-005: Harden trust boundaries for extensions and MCP

• Goal: Make trust expiry, tool access, and extension loading enforceable at call time.
• Scope: MCP trust, skill loading, and subagent isolation risks.
• Dependencies: risk register and governance rules.
• Acceptance Criteria: no expired trust entry continues to act trusted; unsafe extension paths are explicit.
• Tests: trust expiry and permission enforcement regressions.
• Files likely touched: teaagent/mcp_trust.py, teaagent/skill_executor.py, teaagent/subagents/_isolation.py, related tests.
• Risk: high
• Parallelizable: yes
• Human Review Required: yes

TASK-006: Convert the docs corpus into a clearer control plane

• Goal: Keep dated evidence, add supersession notes, and make the shortest path to current truth obvious.
• Scope: analysis indexes, governance docs, and current-status links.
• Dependencies: evidence-to-principle policy.
• Acceptance Criteria: users can find the current truth without reading the whole archive.
• Tests: docs consistency checks and link verification.
• Files likely touched: docs/analysis/daily-driver-review-INDEX-2026-06-01.md, docs/governance/README.md, new evidence docs.
• Risk: medium
• Parallelizable: yes
• Human Review Required: no

TASK-007: Refresh competitor survey and strategic rationale periodically

• Goal: Keep the product rationale aligned with current official docs and community signals.
• Scope: competitor survey, roadmap rationale, critique doc.
• Dependencies: external signal refresh process.
• Acceptance Criteria: the strategy docs state date context and distinguish evidence from inference.
• Tests: manual source verification and link checks.
• Files likely touched: docs/analysis/competitor-signal-survey-2026-06-04.md, docs/strategy/daily-driver-roadmap-rationale-2026-06-04.md, docs/reviews/daily-driver-critique-and-counterarguments-2026-06-04.md.
• Risk: medium
• Parallelizable: yes
• Human Review Required: no

Execution Order

1. Task 3
2. Task 1
3. Task 2
4. Task 4
5. Task 5
6. Task 6
7. Task 7

Status Log

• 2026-06-04 — TASK-004 DONE (Human Review Required: no). First-hour onboarding strengthened: docs/daily-driver-current-status.md gained a "First run (safe)" callout (safe command + where status lives + recovery pointer) and docs/tui-daily-driver-guide.md gained an early "If something goes wrong" pointer to the recovery guide.
• 2026-06-04 — TASK-006 DONE (Human Review Required: no). Docs control plane clarified: the dated docs/analysis/daily-driver-review-INDEX-2026-06-01.md now carries a supersession banner pointing to the current front door (docs/INDEX.md) and current status; docs/governance/README.md already links the front door.
• 2026-06-04 — TASK-003 VERIFIED DONE (no behavior change). Cost truth was already enforced: ChatSessionController is the single source of truth (get_session_cost), the TUI reads it with a local fallback, and spend accumulates additively. Verify-first regression guard: tests/test_task003_cost_truth.py (no fake zero after real spend; honest zero on free runs; additive accumulation).
• 2026-06-04 — TASK-001 VERIFIED DONE (no behavior change). Both CLI (cli/_handlers/chat_repl.py) and TUI route cost and undo through the same ChatSessionController, so trust semantics are surface-independent. Verify-first regression guard: tests/test_task001_surface_parity.py (identical cost + undo outcomes across surfaces; only the output sink differs).
• 2026-06-04 — TASK-002 VERIFIED DONE (no behavior change). Journal-based undo and checkpoint restore are separated in code, help text (journal-first, checkpoint fallback), cockpit availability signals (has_undo_journal vs has_checkpoint), and completion wording. The selected path is derivable before running from cockpit availability + the documented precedence. Verify-first regression guard: tests/test_task002_undo_honesty.py. Optional non-blocking enhancement: add a single explicit "undo will use: …" pre-run line; deferred (would be a user-facing change requiring review).
• 2026-06-04 — TASK-005 VERIFIED DONE (no behavior change). Trust expiry is enforced at call time: merged_tool_filters drops expired servers' tools and the registered pre-tool hook raises an explicit HookError (naming the server) when an expired server's tool is invoked. Verify-first regression guard: tests/test_task005_trust_expiry_enforcement.py.
• Net: all four Human-Review-Required behavior tasks were already satisfied by existing code; the verify-first pass added regression guards only, with no behavior change, so no human-review gate was triggered.
• 2026-06-04 — TICKET-15 DONE. Removed redundant audit_trail field from agent_review.py review JSON (replaced with governance-record comment per previous _agent.py pattern). Removed audit_trail injection from test suspension fixtures and updated review-data assertions to check mode instead. Verified: 203 affected tests pass, full suite 3376 pass.
• 2026-06-04 — TICKET-16 Phase 2 VERIFIED DONE (Option A: run_started event written at suspend time). Roundtrip test test_repl_suspend_resume_roundtrip asserts task_for_run finds the run after suspension. All 5 suspend/resume tests pass.
• 2026-06-04 — TASK-DD2 compliance audit DONE. Cross-referenced 3 standards docs (tool-development.md, integration-guide.md, approval-policy-design.md) against live code. Updated each checklist with pass/warn annotations and per-file evidence references. All 14 TASK-DD2 items verified committed.
• 2026-06-06 — TASK-007 DONE. Competitor survey refreshed to 2026-06-06: scripts/refresh_agent_readme_survey.md, new docs/analysis/competitor-signal-survey-2026-06-06.md, strategy rationale and critique refresh notes, and synchronized landscape survey dates across use-case matrix, catalog, architecture, and use-cases docs via refresh_competitive_docs.py.
• 2026-06-04 — COMPREHENSIVE VERIFICATION INVESTIGATION DONE. Cross-referenced every claimed-done task against actual code, tests, and git history. Summary:
  • TASK-001/002/003/005: 4 regression guard test files verified — exist, substantive, match claims exactly (84–97 lines each, 2–5 tests per file).
  • TASK-004/006: Doc changes verified — "First run (safe)" callout in daily-driver-current-status.md, "If something goes wrong" in tui-daily-driver-guide.md, supersession banner in daily-driver-review-INDEX, governance/README.md links front door.
  • TICKET-12/13/14/15/16 Phase 1+2: All 6 items FIXED — commits found for each, file changes match claims, key code patterns verified, tests exist.
  • TASK-DD2-001–014: 11 clean Fixed, 3 with plan-status discrepancies (005: plan says "Active" but index says "Fixed", git_sandbox.py untouched; 013: plan says "Active support ticket" but headless TUI tests exist; 014: plan says "Active support ticket" but docs synchronized). Plan file status headers need reconciliation.
  • DOCOPT-001–012: 14/14 items pass verification (13 fully, 1 minor linking gap in acceptance.md → coverage omit ledger). ADR statuses confirmed (zero "Proposed" stale entries). Dependency audit lanes verified in CI workflows.
  • TASK-007: Still open (competitor survey — docs/research, not code). Minor gap found: docs/acceptance.md does not explicitly link the coverage omit ledger at docs/governance/coverage-omit-ledger.md. All other claims verified.
• 2026-06-04 — Leaked-item closure + re-verification. The minor gap above is now fixed: docs/acceptance.md "Current Status" section links the coverage omit ledger (governance/coverage-omit-ledger.md). The three TASK-DD2 plan status headers flagged above are also already reconciled — 005 reads "Partially Fixed" (matches index; broader git-sandbox ACs honestly open behind a Human Review gate), 013/014 read "Fixed". Re-verified on Python 3.12.8: full suite 3415 passed, 22 skipped, 0 failed; scripts/validate_docs_consistency.py passes; ruff check and ruff format --check clean. No code task remains open except the intentionally-deferred backlog (TASK-007 competitor survey; DOW-014/018–030 P2 docs-tooling; TASK-DD2-005 broader git-sandbox ACs).