fix(plan): add tier hydration and improve architecture review output #10938

Merged

brent.edwards merged 32 commits from tdd/m3-actor-run-response into master 2026-05-19 12:43:35 +00:00

Conversation 41 Commits 32 Files Changed 68

+2663 -261

32 Commits

Author	SHA1	Message	Date
brent.edwards	bee15529c7	style(format): fix format errors ... CLOSES ISSUE: #10878	2026-05-19 12:22:12 +00:00
brent.edwards	1e21bcb8cd	fix(plan): resolve review issues from PR #10938 comment 272179 ... - C1: Fix _NEW_START/_NEW_END delimiter constants to use 8 trailing > chars matching the LLM prompt template (was 9, causing all file blocks to be silently discarded at parse time — total functional regression) - C2: Move os.makedirs inside OSError handler and guard against empty dirname (flat filenames like "FILE: report.md" would crash execute phase) - C3: Update read-only sandbox test to use CLEVERAGENTS sentinel delimiters instead of backtick format (test was passing vacuously — no files parsed) - M1: Scope tier hydration cache by plan_id using _hydrated_plan_ids set in PlanExecutor to prevent cross-plan contamination where Plan B (linked to Project Y) would skip hydration due to Plan A (Project X) results - M6: Patch hydrate_tiers_for_plan in tier hydration test scenarios for direct assertion on mock.called instead of tautological get_hot_fragments checks; pre-populate _hydrated_plan_ids for the skip scenario - M7: Add Behave scenarios for <CAFS>/<CAFE> short-form and >>>>>>>> arrow delimiter formats which had zero test coverage - M8: Add _ensure_gitignore_entry helper; auto-add plan-output/ to .gitignore when sandbox directory is created to prevent accidental VCS commits of generated files - Nit: Fix CLEVERAGENTS_FILE.End typo -> CLEVERAGENTS_FILE_END in feature file and matching step definition - Nit: Add @tdd @tdd_issue @tdd_issue_10938 tags to tier hydration feature - Nit: Add @tdd @tdd_issue @tdd_issue_10878 tag line to llm_file_parsing regression feature ISSUES CLOSED: #10878	2026-05-19 06:05:11 +00:00
brent.edwards	79436174ee	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-18 14:14:57 -07:00
brent.edwards	8ee00bd677	style(plan_executor_tier_hydration_steps.py): ruff format needed a change ... ISSUES CLOSED: #10878	2026-05-18 20:30:21 +00:00
brent.edwards	a1bee8f79c	fix: resolve code review issues from comment 9244 (PR #10938) ... Fix C1: Align delimiter padding (8 '>' chars) in LLM file-block parser with prompt template. Add NOTE comment to prevent future drift. Fix M1: Set max_tokens on LLM instance for non-configurable providers (anthropic, google, etc.) before invoke() so provider.max_tokens applies. Fix M2: get_context_summary() returns None instead of placeholder string, preventing meaningless text from entering strategy prompts. Fix M3: Add output-path message after plan execution completes. Fix plan.plan_id → plan.identity.plan_id (type error). Fix M5: Assert exact context_max_tokens_hot == 32000 per spec. Fix M6: Fix tier hydration step definitions - use StrategyDecision Pydantic model, store patcher.mock for called-check, fix patch cleanup, add non-empty decisions assertion, delete defunct KeyError scenario. Fix M7: Wrap shutil.copy2 in try/except OSError with warning log. Fix m1: Remove KeyError and RuntimeError from exception handlers in plan_executor tier hydration and strategy_actor tier service context. Fix m2: Guard os.path.relpath() with try/except ValueError for Windows cross-drive paths. Fix m6: Remove stale TODO comment from estimation_actor_steps. Also fix lint error (line too long) in plan.py output message.	2026-05-18 20:18:39 +00:00
brent.edwards	4454e7580b	fix(tests): update test steps for commit_plan rename and _parse_file_blocks tuple return ... - Update feature files: _commit_plan → commit_plan in step text (plan_apply_service_branch_coverage.feature, plan_executor_coverage.feature) - Unpack (entries, blocks) tuple from _parse_file_blocks in all test step callers (llm_actors_coverage_steps, llm_delimiter_regression_steps, llm_file_parsing_regression_steps) - Fix _write_to_sandbox call sites to pass blocks (not entries) and drop the now-removed third llm_output argument - Update merge_conflict_abort_steps: create flat-copy sandbox at plan-output/<plan_id>/ instead of .cleveragents/sandbox/ - Fix automation_profile_cli.feature assertion to match Click error format ('--automation-level' substring check)	2026-05-17 18:23:30 +00:00
brent.edwards	c73df0e1ca	fix(plan): add tier hydration and improve architecture review output ... Implement review comment fixes from PR #10938: HIGH SEVERITY: - H1 (plan.py): Fix stale .cleveragents/sandbox paths to plan-output/<plan_id> in 4 locations (docstring, guard path, fallback docstring, fallback path + _skip_dirs) - H2 (llm_actors.py): Store file content in entry.metadata["content"] for downstream consumers - H3 (llveragents.py): Single-pass parsing - _parse_file_blocks returns (entries, blocks), _write_to_sandbox consumes blocks directly, eliminating double regex parsing - H1 (strategy_actor.py): Remove AttributeError from exception handler (masks programming bugs) MEDIUM SEVERITY: - M3 (llm_actors.py): Remove duplicate tier hydration block from LLMExecuteActor.execute() (hydration already runs in PlanExecutor.run_strategize()) - M4 (llm_actors.py): Remove dead entries parameter from _write_to_sandbox (no longer used) ARCHITECTURE: - M5 (plan_executor.py + llm_actors.py): Rename _commit_plan to commit_plan, add to PlanLifecycleProtocol, update 137 call sites across src/, benchmarks/, features/, robot/ LOW SEVERITY: - L1 (context_tiers_steps.py): Fix hardcoded 32000 assertion to threshold-based (>= 16000) - L2 (strategy_actor.py): Extract magic number 20 to _MAX_HOT_CONTEXT_FRAGMENTS constant Closes #10938 --- Automated by CleverAgents Bot Supervisor: Implementation | Agent: task-implementor	2026-05-17 08:45:55 +00:00
brent.edwards	bdfa791397	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-16 19:38:06 +00:00
brent.edwards	9dbb583da8	fix(plan): address PR review comments — delimiter collision, type safety, and execution metrics ... Address blocking, critical, and high-priority review feedback from PR #10938: CRITICAL: - C1: Fix git merge-conflict marker collision by changing file delimiters from <<<<<<< CLEVERAGENTS_FILE_START >>>>>>> to >>>>>>>> CLEVERAGENTS_FILE_START >>>>>>>> (avoids <<<<<<< prefix that collides with git conflict markers). Prompt updated to emit new delimiters; _parse_file_blocks and _write_to_sandbox both support all three formats: <CAFS>/</CAFE>, >>>>>>> markers, and legacy <<<<<<< markers for backward compatibility. - C2: _write_to_sandbox now supports <CAFS>/</CAFE> delimiters with negative-lookbehind escape support, matching _parse_file_blocks. HIGH: - H1: Remove AttributeError from tier hydration exception handlers in plan_executor.py and strategy_actor.py — AttributeError indicates a programming defect and must propagate rather than be silently swallowed. - H2: Add execution_duration_ms to LLMExecuteActor.execute() return value using time.monotonic_ns() measurement bracketing the LLM call + parse pipeline. - H3: Add decision_ids_processed to LLMExecuteActor.execute() return value (was defaulting to empty list in actor mode). MEDIUM: - M1 (type safety): Change tier_service parameter in LLMExecuteActor from untyped Any to ContextTierService | None via TYPE_CHECKING import to support Pyright strict mode. Testing: All relevant Behave test suites pass (llm_actors_coverage, llm_delimiter_regression, llm_file_parsing_regression, plan_executor_tier_hydration, context_tiers, context_tier_hydration, strategy_actor_llm — 202 scenarios, 635 steps). Refs: PR #10938	2026-05-15 23:18:43 +00:00
brent.edwards	3aa296e301	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-15 14:26:32 +00:00
brent.edwards	5f6bd0b6dc	fix(plan): output plan results ... Fix one more format check. Grump! ISSUES CLOSED: #10878	2026-05-14 22:35:48 -07:00
brent.edwards	0bcade4617	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-15 04:46:17 +00:00
brent.edwards	da415dd7dc	fix(plan): output plan results ... Added the lint fix. ISSUES CLOSED: #10878	2026-05-14 21:42:57 -07:00
brent.edwards	a784e7c9e9	test: fix step pattern mismatches causing CI unit_tests failures #10938 ... main_error_paths_steps.py: - Change `{text:s}` to `{text}` in 'the main cli output contains' step (:s in parse only matches non-whitespace; step text contains spaces) - Change 'result is {expect}' to 'result is "{expect}"' so parse extracts the number without surrounding quotes (was causing assert Expected '"0"', got '0') plan_apply_render_steps.py: - Add singular 'added decision' variant (feature uses singular not plural) - Add 'the patch preview should include the artifacts path' (no arg form) - Add 'the output should contain the mode value "{value}"' (word order) - Add 'the output should contain "{text}" timestamp' variant - Add 'the output should NOT contain a "{text}" line' (with article 'a') - Add '"{text}" mode color markup' and 'rich green checkmark' steps - Add two Given steps for no-guidance with unquoted 'pending execution' transport_selector_steps.py: - Add specific step for empty-string URL ('server_url ""') as parse's {url} requires at least one character and cannot match an empty string --- Automated by CleverAgents Bot Supervisor: PR Fix | Agent: task-implementor	2026-05-15 04:18:21 +00:00
brent.edwards	a23a773168	test(plan): add missing step definitions for delimiter regression tests #10938 ... Add features/steps/llm_delimiter_regression_steps.py with all step definitions for the 8 scenarios in llm_delimiter_regression.feature that were erroring due to missing implementation. Add @tdd_issue tag to llm_delimiter_regression.feature Feature-level tags so the hook validation passes (each @tdd_issue_10878 scenario must also carry @tdd_issue per project policy). Add three missing step definitions to llm_file_parsing_regression_steps.py: - @given('a mock plan_id "01HQESCTEST"') - Given 'an LLM response with a single FILE block using legacy markers where the body contains a backslash-escaped ... sequence' - @when('I parse file blocks using the new delimiter pattern') (short alias) These steps were needed by the new escape-support scenario added in the previous commit (fix(plan): escape delimiter markers #10938). --- Automated by CleverAgents Bot Supervisor: PR Fix | Agent: task-implementor	2026-05-15 03:31:26 +00:00
brent.edwards	6f44583a91	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-14 19:00:11 -07:00
brent.edwards	7369486dae	fix(plan): escape delimiter markers to prevent content collision #10938 ... Resolve CoreRasurae concern (#262743) about delimiter markers conflicting with LLM-generated file content. Changes: - Add negative lookbehind (?<\\!\\\\) to regex patterns so escaped delimiters (backslash-prefixed) in file body are treated as literal text - Use new compact <CAFS>/</CAFE> markers for prompt while maintaining backward compatibility with legacy CLEVERAGENTS markers - Update _parse_file_blocks with escape-aware parsing and path deduplication - Update _write_to_sandbox to unescape delimiter markers in file content before writing - Replace bare Exception handlers with specific exception types (OSError, UnicodeDecodeError, ValueError) --- Automated by CleverAgents Bot Supervisor: PR Fix | Agent: task-implementor	2026-05-15 01:25:48 +00:00
brent.edwards	c927344807	fix(plan): output plan results ... Fix formatting errors. ISSUES CLOSED: #10878	2026-05-14 11:32:25 -07:00
brent.edwards	678c08ca09	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-14 18:26:27 +00:00
brent.edwards	86d6ae6aa3	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-13 22:45:51 -07:00
brent.edwards	938cb3bb1f	fix(plan): output plan results ... Added missing TDD tests. ISSUES CLOSED: #10878	2026-05-13 22:43:40 -07:00
brent.edwards	bf55f9e7d5	fix(plan): output plan results ... Fixing the `ruff check` errors. ISSUES CLOSED: #10878	2026-05-14 04:22:49 +00:00
brent.edwards	6a922719ef	fix(plan): output plan results ... Fixing the code coverage. ISSUES CLOSED: #10878	2026-05-13 19:00:35 -07:00
brent.edwards	5722457b65	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-13 14:29:22 +00:00
brent.edwards	c016fc7b43	fix(plan): output plan results ... ISSUES CLOSED: #10878	2026-05-12 22:33:03 -07:00
brent.edwards	755b610321	fix(plan): additional improvements for context hydration and sandbox management ... Includes more specific exception handling in strategy_actor.py and improved sandbox path resolution in plan.py ISSUES CLOSED: #10878	2026-05-13 02:29:27 +00:00
brent.edwards	73c8c6baa1	fix(plan): output plan results ... This commit addresses the issues with hardcoded max_tokens values and provider compatibility. ISSUES CLOSED: #10878	2026-05-13 02:28:52 +00:00
brent.edwards	5723603ef6	fix(plan): output plan results ... ISSUES CLOSED: #10878	2026-05-12 23:06:16 +00:00
brent.edwards	c557464485	Merge branch 'master' into tdd/m3-actor-run-response	2026-05-12 15:48:18 -07:00
brent.edwards	723bd291e8	fix(plan): resolve TDD branch-introduced test failures ... - Fix WF10 Batch add/add conflicts by using full plan_id instead of plan_id[:8] for plan-output/ sandbox directories, eliminating cross-plan file collisions in batch operations. - Fix WF10 Batch provider crash by catching ValueError from ProviderRegistry.create_llm() in LLMExecuteActor.execute() and returning a no-op ExecuteResult instead of propagating an unhandled exception. - Fix M6 E2E rc=1 by wrapping tier_service.get_hot_fragments() inside the existing hydration try/except block in PlanExecutor.run_strategize(), so missing context-tier database tables cause a graceful fallback instead of a fatal crash.	2026-05-06 23:39:49 +00:00
brent.edwards	d8147bf6f6	fix(plan): output plan results ... Fixes: #10878	2026-05-06 22:35:38 +00:00
brent.edwards	9f6f2148f4	fix(plan): add tier hydration and improve architecture review output ... - Add tier hydration before strategize phase in plan_executor.py - Increase max_tokens to 16384 in llm_actors.py for longer outputs - Change file delimiters from ``` to >>>>>>>/<<<<<<< to avoid Markdown conflicts - Increase context_max_tokens_hot from 16000 to 32000 in settings.py - Fix get_hot_view → get_hot_fragments in strategy_actor.py and plan_executor.py - Add opencode to skip directories in context_tier_hydrator.py - Change sandbox output location to plan-output/ directory in plan.py - Add get_context_summary stub method to acms_service.py This fixes issue #10878 where architecture reviews were truncated because the regex pattern for parsing file output would stop at the first ``` encountered in the Markdown report.	2026-05-06 22:35:35 +00:00