deviatdd 2.20.2__tar.gz → 2.21.2__tar.gz

{deviatdd-2.20.2 → deviatdd-2.21.2}/.deviate/config.toml

@@ -1,5 +1,5 @@
 profile = "default"
-timeout_seconds = 300
+timeout_seconds = 1800
 agent_export_mode = "local"
 
 graphite = false
@@ -7,3 +7,10 @@ use_context = true
 use_libref = true
 [agent]
 backend = "pi"
+transport = "cli"
+pi_rpc = false
+timeout = 1800
+
+[models]
+default = "grok-4.6"
+

{deviatdd-2.20.2 → deviatdd-2.21.2}/.gitignore

@@ -12,6 +12,7 @@ __pycache__/
 .tdd-session.json
 .deviate/session.json
 .opencode/index/
+.codebase-index/
 .opencode/codebase-index.json
 node_modules/
 

{deviatdd-2.20.2 → deviatdd-2.21.2}/CHANGELOG.md

@@ -8,6 +8,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- **`/deviate-pr` now marks the issue COMPLETED before it pushes, and opens a PR/MR only on request.** Rewritten to mirror `/deviate-merge`: `deviate pr run` appends the COMPLETED transition to `specs/issues.jsonl`, stages and commits it with the PR body, pushes the branch, then opens a GitHub PR (`gh pr create`) or a GitLab MR (`git push -o merge_request.create`). Platform is auto-detected from the `origin` remote hostname. `--no-pr` pushes and marks COMPLETED without opening anything; `--merge` / `--auto-merge` apply to GitHub only (GitLab warns and opens for review). `--platform github|gitlab` forces the platform. Pinned by `tests/test_integration/test_meso_layer.py::TestPrRun` and `tests/test_meso/test_pr_platform.py`.
+- **Graphite CLI integration is removed from the active code, prompts, and specs.** The `--graphite` setup flag, the `graphite` config field and `resolve_graphite_config`, the `gt create` / `gt submit` routing in `feature.py` / `micro.py` / `meso.py`, and the `## Graphite Stacked Changes Workflow` governance seed are gone. `deviate pr` no longer has a Graphite path (`git push -o merge_request.create` covers GitLab). Historical spec archives (`specs/adhoc/007-graphite-cli`, `specs/003-graphite-cli-integration`) and the append-only ledgers are left untouched.
+- **Manual slash-command bodies now derive from the canonical `auto/{phase}.md` core at install time.** `install_command`/`compose_command_body` (`src/deviate/core/commands.py`) splice the canonical `auto/{phase}.md` middle verbatim for the 11 overlapping phases (explore, research, prd, shard, plan, tasks, red, green, refactor, judge, execute) plus a per-phase manual overlay (pre/post-script lifecycle, rich handover manifest, `<context><user_input>`). The hand-maintained duplicate middle bodies were deleted from `commands/deviate-{phase}.md`, so the manual and auto modes can no longer drift. The derived RED inherits auto's `status: "PASS"` + `failure_kind` semantics (no `status: "FAIL"`/abort-on-passing-test); the derived GREEN inherits auto's "write ONLY production code" role language. A drift guard pins the identical-middle invariant across all 11 phases. Pinned by `tests/test_core/test_commands.py::TestManualDerivationFromAutoCore` and `tests/test_meso/test_auto_prompt_templates.py::TestManualDerivationDriftGuard`.
 - **The `deviatdd` skill now prepares Meso before it drains Micro.** It first runs idempotent `deviate meso run`, then runs bare `deviate micro run` per task. Inside a feature worktree, Meso skips valid completed phases. Valid Plan plus Tasks emits `MESO_ALREADY_COMPLETE`. Valid Plan without Tasks resumes at Tasks. Missing artifacts run Plan then Tasks. Invalid existing Plan or empty Tasks stop without overwrite. The separate `deviatdd-meso` skill was removed, so `deviate setup` continues to install one unified skill. Pinned by `tests/test_meso/test_meso_resume.py` and the setup skill tests.
 - **Acceptance-contract scenarios now require exactly one `**Verification Mode**` line.** Each `AC-PLAN-NNN` scenario in `plan.md` must carry a `**Verification Mode**: automated|manual|deferred` literal (case-insensitive). `validate_acceptance_contract` (`src/deviate/core/validation.py`) reports a named error — `missing Verification Mode`, `duplicate Verification Mode lines`, or `invalid Verification Mode '<value>'; expected one of automated|manual|deferred` — and `deviate plan post` / `deviate meso tasks pre` block with `PLAN_ACCEPTANCE_CONTRACT_INVALID`; meso-run resume blocks with `MESO_PLAN_INVALID`. Manual/deferred modes validate with no `test_ref` in the body. Pinned by `tests/test_core/test_validation.py::TestVerificationModeValidation` and the gate/resume regression tests in `tests/test_cli/test_meso_contracts.py` / `tests/test_meso/test_meso_resume.py`.
 - **`deviate micro run --auto` spawns the configured agent with the deviatdd skill slash command as the prompt.** Skips the normal RED/GREEN/JUDGE/REFACTOR orchestration and hands the queue off to the agent, which then drives `deviate micro run` per task until the queue drains. Agent-agnostic: the new `_DEVIATDD_SLASH_COMMAND` map resolves the slash command by backend — `/deviatdd` for Claude Code, `/skills:deviatdd` for Pi and OMP. Falls back to `AUTO_NO_SLASH_COMMAND` + exit 1 when the configured backend (e.g. `opencode`, `droid`, `stub`) has no documented slash command form. Pinned by `tests/test_micro/test_run.py::TestAutoFlag` (4 tests: slash command per backend, unknown backend rejection, normal dispatch skipped).
@@ -18,11 +21,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Constitution is now prepended to every phase prompt on both the auto and manual paths.** Closes the gap where agents running via `/deviate-*` slash commands never saw the constitution at the top of the prompt and could silently substitute a mandated tech-stack component (e.g., deferring Phoenix LiveView for a framework-free shell — a `lib/meepleinn_web/live/*.ex` module with a moduledoc disclaimer naming the missing LiveView transport for "future wiring"). `compose_command_body(raw, core_dir, constitution_path=...)` (`src/deviate/core/commands.py`) now accepts a constitution path and prepends the file's content as the first tier of the composed body — same parity as the auto path's `deviate.prompts.assembly.load_template`. `install_command(...)` derives `<workdir>/specs/constitution.md` from the workdir argument (already threaded through `_install_commands_to_agents`) and passes it through. Missing constitution is non-fatal (greenfield case: `deviate setup` runs before `deviate research` scaffolds the constitution). The universal invariant `core.md` now carries an explicit `Constitution Compliance Mandate` (#10) that forbids silent omission, deferral, or substitution of a mandated component without an ADR + `constitution.md` amendment + HITL sign-off — a code comment, issue bullet, or GREEN-phase hand-wave is no longer sufficient. Pinned by `tests/test_core/test_commands.py::TestComposeCommandBodyConstitutionInjection` (3 tests: prepended when path provided, no injection when path is None, no injection when file missing) and `tests/test_core/test_commands.py::TestPlatformFrontmatter::test_installed_command_includes_constitution_when_workdir_has_one` / `::test_installed_command_strips_constitution_when_no_constitution_in_workdir`.
 - **JUDGE phase now treats "Constitution Compliance Violation" as a blocking category.** Adds category #7 (`src/deviate/prompts/auto/judge.md`) and a matching `Constitution Compliance | Critical` evaluation dimension. A diff that substitutes, defers, mocks away, or omits a component the constitution mandates (tech stack, transport, architectural boundary, runtime, framework) is now `COMPLIANCE_VIOLATION` with severity `CRITICAL` — even when the AC-NN's surface behavior appears satisfied and the test passes via a stand-in substrate (socket-shaped map asserting on `:assigns` while the real Phoenix LiveView WebSocket is unwired, REST shim around a LiveView contract, mocked system-under-test to bypass the real transport). STEP_2 of the execution sequence adds a `Constitution scan` hard check that cross-references the constitution (prepended to the JUDGE prompt at tier 1) against the diff. The verdict YAML schema gains `constitution_compliance: "PASS" | "FAIL"`. The "seven Categories" → "eight Categories" tally is updated in STEP_3 and the constraints block. This is a prompt-only change — no production code or test infrastructure changes — so it does not add a runtime regression test; the change is locked by manual prompt review against the MeepleInn/Phoenix scenario from the issue.
 
-- **RED phase that produces no failing test now routes the decision to JUDGE instead of dying in a raw `PhaseFailedError`.** When a task's test command exits 0 (all tests pass) or collects no tests, `deviate micro run` skips GREEN (a vacuous GREEN had nothing to implement and burned the TRAIN budget) and calls JUDGE directly with a new `failure_kind: no_failing_test` discriminator. JUDGE either rules the required behavior already exists (`skip_refactor` / bare `COMPLIANCE_PASS` → task marked COMPLETED, the uncommitted passing test discarded) or rules the test wrong (`revert_before` → RED re-authors a genuinely failing test). JUDGE's diff spans the uncommitted RED test so it can review what the agent actually wrote. The RED agent may declare `failure_kind: already_satisfied` / `test_defect` with a `rationale` to steer the adjudication. Pinned by `tests/test_micro/test_orchestration.py::TestMicroOrchestration::test_micro_red_no_failing_test_routes_to_judge_skip_refactor`.
+- **RED phase that produces no failing test now routes the decision to JUDGE instead of dying in a raw `PhaseFailedError`.** When a task's test command exits 0 (all tests pass) or collects no tests, `deviate micro run` skips GREEN (a vacuous GREEN had nothing to implement and burned the TRAIN budget) and calls JUDGE directly with a new `failure_kind: no_failing_test` discriminator. JUDGE either rules the required behavior already exists (`skip_refactor` / bare `COMPLIANCE_PASS` → task marked COMPLETED, declared regression tests stay on disk) or rules the test wrong (`revert_before` → RED re-authors a genuinely failing test). JUDGE's diff spans the uncommitted RED test so it can review what the agent actually wrote. The RED agent may declare `failure_kind: already_satisfied` / `test_defect` with a `rationale` to steer the adjudication. Pinned by `tests/test_micro/test_orchestration.py::TestMicroOrchestration::test_micro_red_no_failing_test_routes_to_judge_skip_refactor`.
 - **`deviate specify --local` claims an issue without pushing to the remote.** When `--local` is set, `_try_claim_issue` skips the `branch_exists_on_remote` check and the `git push` step, so the worktree is created, the ledger row is written, and the claim commit is recorded locally only. If the local branch `feat/<epic>/<slug>` already exists, the command short-circuits with `ALREADY_CLAIMED_LOCAL` and reuses the existing worktree, treating branch presence as the claim signal without rewriting the ledger. NOTE: this semantic can false-positive on a manual `git checkout -b feat/<epic>/<slug>` that pre-dated any claim — there is no opt-in to disable it for the no-remote workflow, so prefer `deviate specify <id>` (the default, which checks remote first) when operating against a shared origin. The new option lives at `src/deviate/cli/meso.py::_try_claim_issue(..., local=...)` and the `specify` Typer command forwards it through `_specify_pre`. Pinned by `tests/test_meso/test_specify.py::TestSpecifyLocalFlag` (2 tests: branch-exists short-circuit, kwargs forwarded) and `tests/test_cli/test_meso.py::TestSpecifyLocalFlag` (4 tests: remote check skipped, push skipped, short-circuit dict shape, branch-exists short-circuit).
 - **`deviate flows sync` — sole owner of `specs/_product/flows.jsonl` creation.** New top-level `flows` Typer group exposes a single `sync` subcommand that parses `specs/_product/flows/index.md` and appends one ``FlowRecord`` identity row plus ``FLOW_DISCOVERED`` and ``FLOW_DOCUMENTED`` events per flow. Re-running on a populated ledger is a no-op (compound-key idempotency on the underlying append helpers). Exits non-zero with ``FLOWS_INDEX_MISSING`` on stderr when the canonical index is absent, and ``FLOWS_INDEX_EMPTY`` when the index parses to zero rows — surfaces authoring defects instead of silently committing a half-baked catalog. New public service `seed_flow_ledger(flows_index, ledger_path)` in `src/deviate/state/ledger.py` plus `FlowIndexEmptyError` for the empty-index contract. `deviate explore post` no longer seeds identity or documentation events; it only renders the coverage report. ``flows.jsonl`` never receives ``FLOW_REFERENCED_BY_ISSUE`` events — the referenced-by relationship is derived read-only from ``specs/issues.jsonl::flow_refs`` at coverage time.
 - **`deviate html prd --bucket <slug>` targets a specific epic when more than one owns a `prd.md`.** Previously the command hard-failed with `HTML_AMBIGUOUS_PRD` whenever multiple numbered epics had a `prd.md` and offered no flag to disambiguate — the agent was told to "run from within the epic's worktree", which silently yielded `HTML_NO_PRD` because the resolver reads `specs/` from the repo root. The new option resolves `specs/<bucket>/prd.md` directly and bypasses ambiguity detection; the plain form keeps its existing behavior but its ambiguity banner now points at the flag. An unknown/absent bucket exits `PRD_NOT_FOUND`. The `/deviate-html` prompt (`src/deviate/prompts/commands/deviate-html.md`) and spec docs now reference `--bucket` instead of the broken cwd fallback. Pinned by `tests/test_cli/test_html.py::test_html_prd_bucket_targets_specific_epic`, `::test_html_prd_bucket_targets_unnumbered_dir`, and `::test_html_prd_bucket_missing_file_exits_cleanly`.
+- **Optional push-as-lock: `claim_remote` config plus `--local` on `meso run` and `run`.** Standing `.deviate/config.toml` key `claim_remote` defaults to `true` (absent file or absent key still push). `deviate setup --no-claim-remote` writes `claim_remote = false` without dropping `[models]`, `timeout_seconds`, or `[agent]`. Fresh setup without the flag writes `true`. Effective local mode is `--local` OR `claim_remote = false`; explicit `--local` always wins. Local mode still creates `.worktrees/feat/{epic}/{issue}/`, writes SPECIFIED, and commits the claim, and skips `branch_exists_on_remote` plus `git push`. `deviate specify --local` stays; omitted `--local` now honors config. `deviate meso run --local` and `deviate run --local` share that meaning. `--no-setup` remains a distinct skip of worktree plus claim. Local discovery does not treat an origin branch as claimed-elsewhere. Specs: `specs/DeviaTDD-api.md`, `specs/DeviaTDD-architecture.md` (Atomic Concurrency Protocol: push-as-lock is the default, not mandatory). Pinned by `tests/test_state/test_config.py`, `tests/test_cli/test_meso.py`, `tests/test_meso/test_specify.py`, `tests/test_meso/test_meso_orchestration.py`, `tests/test_cli/test_init.py`, and `tests/test_cli/test_top_level_run.py`.
 ### Changed
+- **`deviate refactor pre` now scopes `files_to_refactor` to the RED+GREEN production set (GH-98).** The command no longer glob()s every `src/**/*.py` or discards `_resolve_task_context`. It lists production files from `HEAD~2..HEAD`, falling back to the task `Files:` list minus tests when that git range is empty or unavailable. Test files are never included. The JSON contract now emits the documented handover fields (`status`, `task_id`, `task_title`, `task_type`, `test_command`, `lint_command`, `spec_dir`, `verification`, `repo_root`, `git_branch`, `timestamp`) alongside `files_to_refactor`. Auto `_build_auto_prompt("refactor")` injects the same scoped list and keeps the `git log -2` / `git diff HEAD~2..HEAD` inspect step. Pinned by `tests/test_micro/test_refactor.py`.
+- **GREEN / REFACTOR / review prompts fold smallest-change into existing lines** (reuse stdlib or an already-installed dep; in-place refactor; Opportunities do not extract helpers). Prompt-only. Pinned by `tests/test_meso/test_auto_prompt_templates.py::TestSmallestChangeFoldedIntoExistingPrompts`.
+- **JUDGE evidence is task-scoped; `deviate review` fail-closes on an unclaimed plan AC.** TDD `_run_judge_phase` resolves required `AC-PLAN-NNN` tokens via `resolve_task_ac_tokens` (non-empty `acceptance_criteria` `criterion_id`s, else this task's `tasks.md` card, else none) and does not fall back to every token in `plan.md`. ISS-ADH-020 exact-substring, path-in-diff, and uniqueness-floor checks still apply to that set. EXECUTE, IMMEDIATE, and DIRECT stay ungated. `deviate review pre` / `post` run a runner-owned Gate 3 scan (`evaluate_review_coverage`) and exit non-zero with `COVERAGE_INCOMPLETE` when any this-issue plan token has no COMPLETED claim. Missing `plan.md` or missing plan tokens stay vacuously READY. PENDING, FAILED, and sibling-issue rows do not claim. Pinned by `tests/test_core/test_judge_evidence.py`, `tests/test_micro/test_judge.py`, and `tests/test_cli/test_review.py`.
+- **TDD JUDGE `COMPLIANCE_PASS` requires mechanical diff evidence.** `_run_judge_phase` fail-closes unmatched PASS: missing, empty, or partial `HandoverManifest.evidence`, hallucinated paths, or quotes that are not exact substrings of the injected RED+GREEN `<diff>` rewrite the action to `revert_to_red` with runner-authored feedback. Unmatched PASS cannot COMPLETE. Auto and manual judge prompts require the `evidence` schema and no longer instruct default-pass. EXECUTE and IMMEDIATE judge stay ungated. Pinned by `tests/test_core/test_judge_evidence.py` and `tests/test_micro/test_judge.py`.
+- **TDD `revert_to_red` is fatal when the RED boundary SHA is missing.** JUDGE no longer prints `ROLLBACK_FAILED` and then trains GREEN. `_run_judge_phase` re-raises `PhaseFailedError` carrying `ROLLBACK_BOUNDARY_MISSING`. `_commit_judge_feedback_and_advance` does not stamp `session.red_commit_sha` onto a `docs(...): add judge feedback` commit when no RED-phase failing-test SHA exists. A real RED SHA still rolls back, advances past the feedback commit, and retries GREEN. Pinned by `tests/test_cli/test_micro.py::test_revert_to_red_missing_red_commit_sha_is_fatal` and `test_judge_revert_to_red_advances_red_commit_sha`.
+- **Remote-aware issue and epic ordinals.** Next issue and epic `NNN` is `max(ordinals) + 1` over the current ledger, the `origin/<base_branch>` ledger, and already-fetched remote `feat/<epic>/<NNN>-*` / `feat/adhoc/<NNN>-*` refs. `ISS-ADH-NNN` and `ISS-NNN` share one adhoc series. Local-only unpushed feat branches do not reserve. When `git push` of `feat/.../NNN-*` rejects because the name exists, the claim path increments and retries (cap 3) and does not fall back to `--local`. Explicit `--local` / `claim_remote = false` stay the optional skip. Specs: `specs/DeviaTDD-api.md`, `specs/DeviaTDD-architecture.md`. Prompt: `src/deviate/prompts/commands/deviate-adhoc.md`.
+- **TDD retry now uses two persisted counters (`green_attempts` / `red_attempts`) instead of a single in-memory train budget.** GREEN trains three times against one RED contract on `revert_to_red`, then escalates. `revert_before` (including `_coerce_judge_action` for `test_defect` / `no_failing_test`) escalates immediately. Three RED escalates print `TRAIN_EXHAUSTED` and stop the infinite `revert_before` loop. Counters live on `SessionState` in `.deviate/session.json` so `git reset --hard` does not zero the budget.
+- **Shard Pass 1.5 drops the 4–8 floor; tasks 30–90 names one fail-to-pass contract.** Pass 1.5 now emits as few independently shippable user-visible verticals as the PRD needs: 1 is legal, hard ceiling 10 and `SLICE_CAP_EXCEEDED` remain. The tasks 30–90 rule names one observable fail-to-pass contract and forbids fake splits of the same AC (test-skeleton vs implement vs add-the-route). Eight real verticals still emit eight issues; a mixed 10-file / >400 LOC GREEN packet still splits. Prompt/spec-only: `deviate-shard` / `auto/shard`, `deviate-tasks` / `auto/tasks`, `micro-shared`, `auto/refactor`, `specs/DeviaTDD-api.md`, `specs/DeviaTDD-architecture.md`.
 - **Ad-hoc issue creation no longer marks the record COMPLETED at creation time.** `/deviate-adhoc` step 7 previously instructed the agent to “mark the record as completed” via `deviate adhoc post` — but that CLI reads a different ledger (`specs/adhoc.jsonl`) than the `specs/issues.jsonl` ledger the prompt populates, so agents instead appended a raw `COMPLETED` transition to `specs/issues.jsonl` immediately, before any plan/tasks/red-green work ran (visible as instant `BACKLOG → SPECIFIED → COMPLETED` sequences for `ISS-ADH-003/004/013`). Step 7 now commits artifacts with a plain `git commit`, leaves the record at `BACKLOG`, and forbids appending any `COMPLETED` transition at creation — completion is written only by later phase post-scripts. The re-installed mirrors (`.opencode/`, `.claude/`, `.omp/`, `.factory/`) carry the fix via `deviate setup`. Prompt-only change in `src/deviate/prompts/commands/deviate-adhoc.md`; pinned by the existing `TestConsumerRepositoryPromptBoundaries` prompt tests.
 - **`/deviate-tasks` no longer waits for the removed Gate 2 approval.** The prompt still said the workflow “stops for joint human review” and that “TDD begins only after explicit artifact-bound Gate 2 approval” — but constitution 0.8.0 removed Gate 2, so the agent was being told to wait for an approval that no longer exists. The Meso Workflow Position line and the Tasks/TDD bullets now state that no human-approval gate sits between Tasks and Micro and that `deviate run` chains meso into micro end-to-end. Prompt-only change in `src/deviate/prompts/commands/deviate-tasks.md`, consistent with the constitution's Automated Advance invariant.
 - **RED phase now hard-fails when it produces no test files (`RED phase produced no test files`), instead of silently committing a vacuous 'failing test' commit.** Previously `_run_red_phase` skipped its test run entirely when `_find_test_files` returned empty, letting GREEN run against nothing and die in `TRAIN_EXHAUSTED` — a real stall. It now raises a `PhaseFailedError` naming `/deviate-execute` (DIRECT) or `/deviate-meso` re-sharding for tasks that legitimately need no code. It also treats a pytest-style exit code 5 (no tests collected) as a RED defect rather than a passing RED. The manual `deviate red post` `RedMustPassError` message now points the agent at the `failure_kind: already_satisfied` escape hatch instead of leading to an opaque crash. Pinned by the updated `_find_test_files`-mocked RED unit tests.
@@ -38,6 +50,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **`/deviate-merge` no longer auto-pushes after the squash-merge commit; the push gate runs inline and the network push is opt-in.** The slash command previously ran `git push` as its final step. As of v2.4.0 the squash-merge commit lands on `main`, then a new `push_gate` step inlines the body of `.githooks/pre-push` (lint + format-check + testmon-driven affected tests with the warm-cache / full-suite fallback — bash 3.2 portable, `GIT_DIR` reset + trap preserved) so the safety net fires even though no `git push` happens yet. After the gate passes the skill asks the operator whether to `git push` (which fires the real `pre-push` hook and re-runs the same gate) or stop and push manually. The squash-merge commit and the ledger transition inside it are durable on `main` regardless of the push outcome — only the network push is deferred. New failure states: `Push_Gate_Failed` (inline gate non-zero), `Push_Failed` (`git push` non-zero, raw stderr surfaced), `Push_Deferred` (user chose "Stop — I'll push manually"). Inline gate body and `.githooks/pre-push` body must stay byte-equivalent; divergence is pinned by `tests/test_meso/test_auto_prompt_templates.py::TestMergePromptPushGate::test_hook_and_prompt_agree_on_gate_body` (which compares non-blank non-comment lines in both bodies and fails on drift in either direction) plus 3 supporting assertions on the upstream-first logic, the testmon fallback, and the prompt structure. Prompt: `src/deviate/prompts/commands/deviate-merge.md` (v2.3.0 → v2.4.0). Spec mirrors updated: `specs/DeviaTDD-architecture.md` (Merge bullet, new `**Push gate + opt-in push (v2.4.0)**` sub-bullet) and `specs/DeviaTDD-api.md` (new `**/deviate-merge push behavior (v2.4.0)**` entry under the `deviate merge` reference).
 
 ### Fixed
+- **`resolve_task_ac_tokens` no longer treats runner-appended `**Judge Feedback**` as required AC tokens (GH-89).** Card-text fallback still follows first-hit order (`acceptance_criteria` `criterion_id`s, else the task card, else none) but now drops `**Judge Feedback**` bullets and their continuation lines before scanning for `AC-PLAN-NNN`. A feedback line quoting `AC-PLAN-001` does not add that token when the card body does not name it. `_task_card_text` still returns the full card. Pinned by `tests/test_core/test_judge_evidence.py::TestResolveTaskAcTokens::test_judge_feedback_quote_does_not_add_token`.
+- **JUDGE revert-route `train_feedback` no longer persists discarded `file:line` citations (GH-103).** After `revert_before` / `revert_to_red`, `_judge_feedback_from_manifest` strips `path:line` tokens before writing `session.train_feedback` and `tasks.md`. The durable rewrite contract ("The next RED/GREEN attempt must: …") stays. Rollback SHA selection is unchanged. Auto judge prompt also forbids line citations on those routes. Pinned by `tests/test_micro/test_judge.py::TestRevertFeedbackStripsLineCitations`.
+- **`deviate micro run` now honors `[agent].timeout` from the worktree `.deviate/config.toml` (GH-87).** `_invoke_agent` previously constructed `AgentConfig(backend=...)` with the 600s default, so long GREEN runs died with `AGENT_TIMEOUT` even when the operator set `timeout = 1800`. Meso already read the same key. Unset or invalid values still default to 600. Pinned by `tests/test_cli/test_micro.py::TestInvokeAgentConfigTimeout`.
+- **Failing GREEN no longer reaches REFACTOR or COMPLETED (GH-97).** After GREEN implements and the suite stays red, the runner logs `TEST_FAILURE` and may still call JUDGE. Only `revert_to_red` (TRAIN / retry GREEN) and `revert_before` (escalate RED) change the route. `continue_refactor` / `skip_refactor` / `proceed_to_refactor_no_diff` / bare `COMPLIANCE_PASS` remap to TRAIN with the pytest dump. `_finish_tdd_cycle` refuses REFACTOR and COMPLETED while that TEST_FAILURE is in effect. Mechanical / test_defect GREEN `status: FAILURE` routing is unchanged. Detection is the test-dump prefix plus empty `failure_kind`, not "any `train_feedback` on GREEN". Pinned by `tests/test_micro/test_orchestration.py::TestMicroOrchestration::test_green_test_failure_compliance_pass_continue_refactor_retrains`.
+- **`/deviate-merge` squash-merges onto the configured `base_branch`, not a hardcoded `main`.** The skill (`src/deviate/prompts/commands/deviate-merge.md`) now checks out, logs, diffs, and refuses against `{base_branch}` from `deviate merge pre` (`resolve_base_branch` / `.deviate/config.toml` `base_branch`; default still `main` when unset). Repos whose trunk is `develop` / `wb-dev` / anything else no longer squash onto the wrong branch. The ledger-only `deviate merge` run path (`_merge_run`) is unchanged. Specs: `specs/DeviaTDD-api.md`, `specs/DeviaTDD-architecture.md`. Pinned by `tests/test_meso/test_auto_prompt_templates.py::TestMergePromptBaseBranch` and `tests/test_cli/test_merge.py::test_merge_pre_emits_configured_base_branch`.
+- **Hung RED now logs `AGENT_TIMEOUT` inside the harness budget and rolls the worktree back to `red_baseline` (ISS-ADH-027).** A `pi -p` child that writes files or trickles stdout and never returns a handover manifest raises `AgentTimeoutError` at `AgentConfig.timeout` (default 600s). `_invoke_agent` logs `AGENT_TIMEOUT`. `_run_red_phase` calls `_restore_worktree_to_baseline`. The operator does not wait for an outer ~1800s bash kill. Stderr stays diagnostic (ISS-ADH-025). EXECUTE still passes `stall_timeout=3600`. Specs: `specs/DeviaTDD-api.md` Agent Backend Hardening and `specs/DeviaTDD-architecture.md` §10.0.
+- **Default Pi spawn is lean, and schema-limit tokens surface immediately on `deviate micro run` (ISS-ADH-026).** `AgentBackend.invoke` appends `--no-extensions`, `--tools read,bash,edit,write`, and `--no-skills` after the existing `pi -p` or `pi --mode rpc --no-session` prefix, plus `--skill` to `.pi/skills/deviatdd/SKILL.md` when that file exists. The first `tool_count_limit` / `unsupported_tool_schema` line kills the child. `_invoke_agent` logs `AGENT_ERROR` with those tokens. RED, GREEN, and REFACTOR raise `PhaseFailedError` that includes the tokens instead of only `agent returned no manifest`. Specs: `specs/DeviaTDD-api.md` Agent Backend Hardening and `specs/DeviaTDD-architecture.md` §10 Pi backend. Pinned by `tests/test_core/test_agent.py` lean-spawn and schema-abort tests, `tests/core/test_agent.py` RPC lean pins, and `tests/test_cli/test_micro.py::TestSchemaLimitHarnessSurface`.
+- **Hung GREEN now logs `AGENT_TIMEOUT` inside the 900s stall budget (ISS-ADH-025 / GH-61).** Stderr diagnostics no longer reset the streaming stall clock. Only stdout is liveness. A stdout-silent GREEN raises `AgentTimeoutError` (`STALL_DETECTED`) and `_invoke_agent` logs `AGENT_TIMEOUT` without a second 900s retry. EXECUTE still passes `stall_timeout=3600`. Specs: `specs/DeviaTDD-api.md` Agent Backend Hardening and `specs/DeviaTDD-architecture.md` §10.0.
+- **`deviate micro run` now treats a known feature-branch issue as authoritative even when the leftover session id still has a `tasks.md`, and meso writes that id into the worktree session.** A leftover `session.active_issue_id` no longer owns the queue when the current `feat/{bucket}/{slug}` branch maps to a different issue. `_rekey_session_issue_to_branch` prefers the branch issue, persists it to `.deviate/session.json`, and scopes bare `deviate micro run`, `--all`, and pinned `TSK-NNN-NN` lookup to that issue. `NO_PENDING_TASKS` exit 0 remains only when the branch issue itself has no pending tasks. `deviate meso run` claim/copy and `MESO_ALREADY_COMPLETE` rewrite the worktree session to the claimed issue so a copied main-repo id cannot stick. The earlier GH-54 bullet covers the no-board re-key only. Pinned by `tests/test_cli/test_micro.py::TestResolveTaskContextUsesBranch` leftover-with-board tests and `tests/test_meso/test_meso_resume.py` / `tests/test_meso/test_meso_orchestration.py` session-write tests.
+- **Pinned `deviate micro run <task-id>` no longer treats a sibling COMPLETED same-number TSK as `TASK_ALREADY_DONE` when this issue is still PENDING.** `_run_single` prints `TASK_ALREADY_DONE` only when the resolved record belongs to the active branch or re-keyed session issue. A foreign COMPLETED row re-resolves this issue's PENDING pin from `tasks.md` or exits `TASK_NOT_FOUND`. Same-number `TSK-NNN-NN` ids stay a per-issue namespace. Pinned by `tests/test_micro/test_run.py::TestPinnedRunIssueScopedAlreadyDone`.
+- **Already-exists COMPLETE on a test-bearing TDD task requires named present regression tests.** RED `failure_kind: already_satisfied` with null or empty `files` and `test_file` raises `PhaseFailedError` and writes no COMPLETED row. JUDGE `skip_refactor` / bare `COMPLIANCE_PASS` requires every declared path in the injected `<diff>` or HEAD. `_restore_worktree_to_baseline` keeps those paths on disk. EXECUTE / IMMEDIATE / DIRECT stay ungated. Pinned by `tests/test_micro/test_orchestration.py` already_satisfied pins and `tests/test_micro/test_judge.py::test_already_satisfied_declared_files_missing_from_diff_fails`.
+- **`no_failing_test` / `revert_before` re-invokes RED (or fails) and never GREEN without a RED-phase SHA.** After JUDGE `revert_before` / cycle `no_failing_test_adjudicated`, `deviate micro run` dispatches RED or raises `TRAIN_EXHAUSTED` / `PhaseFailedError`. It never invokes the GREEN agent while `session.red_commit_sha` is empty or a `docs(...): add judge feedback` SHA. Missing-boundary `revert_to_red` is fatal: `_run_judge_phase` raises `PhaseFailedError` carrying `ROLLBACK_BOUNDARY_MISSING` and does not print `ROLLBACK_FAILED` or train GREEN. Pinned by `tests/test_micro/test_orchestration.py::test_no_failing_test_revert_before_invokes_red_not_green`, `tests/test_micro/test_two_counter_retry.py::test_escalate_to_red_does_not_dispatch_green_without_red_sha`, `tests/test_micro/test_green.py` GREEN-entry pins, and `tests/test_cli/test_micro.py::test_revert_to_red_missing_red_commit_sha_is_fatal`.
+- **Meso Tasks planner locks `Verification_Batch` to `Mode: IMMEDIATE` (GH-57).** The decision tree could assign `Mode: TDD` (and Green "No production change") to a `Verification_Batch`, so RED's fail-first gate rejected already-passing tests. Type→mode routing now maps `Verification_Batch` → `IMMEDIATE` in the planner prompts (`deviate-tasks.md` / `auto/tasks.md`) and in `resolve_execution_mode` (`src/deviate/core/tasks_ledger.py`), used by `generate_jsonl_from_md` and the micro pending-task scan. Other types still pick TDD vs IMMEDIATE. Pinned by `tests/test_core/test_tasks_ledger.py::TestResolveExecutionMode` and `::test_verification_batch_maps_to_immediate_even_when_mode_is_tdd`.
 - **Language-agnostic test discovery: the RED gate no longer globs Python-only `tests/**/test_*.py`.** `_find_test_files` returned `[]` for non-Python projects, so `_run_red_phase` rejected a correctly authored Elixir test (`test/integration/tailwind_pipeline_test.exs`) with "RED phase produced no test files" before any command ran — blocking every task on an Elixir/Phoenix repo. The RED gate now trusts the resolved test command (`_test_command_candidates`: task `verification` → constitution `test_command` → `mise run test` → manifest table `mix.exs`→`mix test` / `Cargo.toml`→`cargo test` / `go.mod`→`go test ./...` / `package.json`→`npm test` / `pyproject.toml`→`pytest`), and a project with no resolvable command (returncode 127) routes to the same JUDGE adjudication as pytest's exit-5 no-tests case instead of dying. `deviate red post`, `deviate green post`, and `deviate refactor post` switched their `TEST_NOT_FOUND`/`NO_TESTS_TO_CHECK` gates from the Python glob to the same resolved-command check. `_find_test_files` remains only as the Python fallback signal. Pinned by `tests/test_micro/test_run.py::TestRunCommand::test_run_elixir_repo_red_accepts_exs_test` (fails on the old gate, passes after).
 - **Manifest schema recovery now drops out-of-enum fields instead of aborting the phase (GH-52, GH-55).** `AgentBackend.parse_output` re-constructed a `HandoverManifest` with the offending value intact after a `ValidationError`, so an out-of-enum `failure_kind` (RED) or `next_action` (JUDGE COMPLIANCE_PASS) re-raised the identical error unguarded and the phase died with a misleading `agent returned no manifest`. Recovery now deletes every top-level field that failed validation (the model default `None` applies); the JUDGE runner's existing `_coerce_judge_action` then routes an unknown `next_action` through the legacy pass path. Pinned by `test_recovery_drops_invalid_failure_kind` / `test_recovery_drops_invalid_next_action` in `tests/test_core/test_agent.py`.
 - **`deviate micro run` dedupes latest task records by `(issue_id, task_id)` so a same-numbered task in another issue no longer shadows the active issue's ledger.** `_collect_latest_task_records` previously keyed the latest record by task ID alone. Since every issue reuses the same `TSK-NNN-NN` numbering, a later-sorting ledger (e.g. `specs/adhoc/*` after `specs/005-*`) shadowed the active issue's records — bare `deviate micro run` reported `no ledger entry` against a committed ledger and explicit `<tid>` dispatch hit the wrong issue's task. Latest records are now keyed per issue, and `_find_task_record` prefers the record owned by the current worktree's branch issue. Pinned by `test_find_task_record_prefers_branch_issue` in `tests/test_micro/test_e2e.py`.

{deviatdd-2.20.2 → deviatdd-2.21.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: deviatdd
-Version: 2.20.2
+Version: 2.21.2
 Summary: DeviaTDD CLI — agent orchestration framework
 Project-URL: Homepage, https://github.com/wernerbisschoff/deviatdd
 Project-URL: Repository, https://github.com/wernerbisschoff/deviatdd

deviatdd-2.21.2/pr_descriptions/feat/adhoc/017-optional-push-as-lock.md

@@ -0,0 +1,5 @@
+Optional push-as-lock for work repos: standing `claim_remote` in `.deviate/config.toml` (default `true`) plus `--local` on `deviate specify`, `deviate meso run`, and `deviate run`. Flag overrides config. Worktree + `feat/{epic}/{issue}` + ledger SPECIFIED stay; only the remote-branch lock and `git push` are skipped when local.
+
+`deviate setup --no-claim-remote` persists `claim_remote = false`. Personal projects with no config change still push the claim branch.
+
+Closes #64

deviatdd-2.21.2/pr_descriptions/feat-adhoc-017-two-counter-tdd-retry.md

@@ -0,0 +1,5 @@
+`deviate micro` used a single in-memory `train_attempts` budget. JUDGE `revert_before` / `test_defect` zeroed that counter and re-ran RED, so an always-escalate stub looped (181 scratch restarts) instead of handing off. ISS-ADH-017 persists two session counters so GREEN trains three times, then escalates, and three RED escalates print TRAIN_EXHAUSTED.
+
+Session: `SessionState` now stores `green_attempts` / `red_attempts` (default 0) and copies them through `transition_to` / `force_transition_to` via `model_dump()` (`src/deviate/state/config.py`). Loop: `_run_tdd_cycle` seeds from session, trains GREEN on `revert_to_red`, escalates immediately on `revert_before` or GREEN-budget exhaust, injects a short `previous cycle failed because …` note (not the GREEN dump), and stops at 3/3 (`src/deviate/cli/micro.py`). `_coerce_judge_action` and EXECUTE `max_judge_attempts` are unchanged. Specs: API, architecture, and CHANGELOG `[Unreleased]` describe the two-counter contract. Tests: `tests/test_micro/test_two_counter_retry.py` plus orchestration / session / revert_before pins.
+
+Closes #59

{deviatdd-2.20.2 → deviatdd-2.21.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "deviatdd"
-version = "2.20.2"
+version = "2.21.2"
 description = "DeviaTDD CLI — agent orchestration framework"
 readme = "README.md"
 license = "MIT"

deviatdd-2.21.2/scripts/verify_install.py

@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from deviate.core.agent import (
+    MAX_PROMPT_CHARS,
+    STREAM_STALL_TIMEOUT_SECONDS,
+    AgentBackend,
+    MalformedHandoverManifestError,
+)
+
+
+def main() -> None:
+    assert MAX_PROMPT_CHARS == 80_000
+    assert STREAM_STALL_TIMEOUT_SECONDS == 900
+
+    try:
+        AgentBackend.parse_output("task_id: TSK-INSTALL-CHECK\n", "pi")
+        raise AssertionError("bare task_id line should not parse as a manifest")
+    except MalformedHandoverManifestError as exc:
+        assert "No YAML handover manifest" in str(exc)
+
+    escaped_quote_hint = AgentBackend._yaml_error_hint(
+        '```yaml\nphase: "JUDGE"\nstatus: "PASS"\ndetail: "x == \\"y\\""\n```'
+    )
+    assert "Avoid backslash-escaped quotes" in escaped_quote_hint
+
+    recovered = AgentBackend.parse_output(
+        "```yaml\nphase: RED\nstatus: PASS\ntask_id: TSK-INSTALL-CHECK\n```\n",
+        "pi",
+    )
+    assert recovered.phase == "RED"
+    assert recovered.status == "PASS"
+    assert recovered.is_success
+
+    print("INSTALL_VERIFIED")
+
+
+if __name__ == "__main__":
+    main()