remove-graph-depth-guard — WorksGood live mirror

Metadata

Status	done
Assigned	`agent-732`
Created	2026-07-23T08:39:09.582813108+00:00
Started	2026-07-25T13:00:49.403573642+00:00
Completed	2026-07-25T13:32:32.601319802+00:00
Tokens	66686351 in / 97339 out

Description

Objective

Remove the bogus maximum task dependency-depth guard. The graph model must permit chains/subgraphs as deep as users need. A visualization or prompt concern must never reject valid graph structure. When historical depth becomes visually noisy, users can archive completed/old tasks and inspect the remaining induced subgraph(s).

Required behavior

Remove guardrails.max_task_depth as an enforced task-creation constraint from add/edit/insert/rescue/import/federation and all other graph mutation paths. There is no semantic depth maximum.
Remove the depth ceiling from agent prompts and user-facing setup/config/TUI settings. Migrate existing config safely: accept and ignore the obsolete key with a clear one-time/read-only lint notice or migration, without breaking old projects.
Keep cycle validation and resource safety, but express them as bounded/iterative algorithms, total-size/work budgets, cancellation, and asynchronous visualization—not as a limit on valid graph depth.
Ensure graph algorithms that can encounter deep chains are iterative or otherwise stack-safe: load/save, readiness, critical path, dependency closure, placement, archive, reset, why-blocked, impact, federation, HTML/viz derivation, and TUI navigation.
Visualization must handle arbitrarily deep graphs through scroll/virtualization, collapse/filter, and asynchronous bounded derivation. It may simplify presentation under load but may not hide/reject legitimate active tasks.
Archiving old/completed tasks must yield a clear view of the remaining induced subgraph or multiple resulting connected components. Preserve dependency/history integrity and provide honest boundary markers where archived predecessors/successors were removed from the active view; never rewrite history merely to make layout shallow.
Existing archived-task recovery, search, exact IDs, task logs, retries/evaluation, and graph commands remain correct.

Validation

Construct and mutate visible chains substantially deeper than 8 (including at least 1,000 nodes) without configuration overrides, rejection, recursion overflow, or quadratic UI blocking.
Deep-chain readiness, critical path, why-blocked, impact, reset/archive, serialization/reload, service dispatch, and TUI/HTML derivation are correct and bounded; inserting a deep active task updates a running TUI promptly rather than taking tens of seconds or remaining offscreen without a discoverable pulse/jump.
Archive an old middle/prefix region and show the remaining induced component(s) clearly with honest archived-boundary context; unarchive/recovery preserves exact edges/history.
Cycles and malformed graphs still fail or render safely for their own reasons, never because of depth.
Legacy max_task_depth config migrates/ignores safely and disappears from prompts, config UX, and Settings.
Run fmt, clippy, full tests and large/deep graph performance smokes under bounded disk/CPU; never globally install an unmerged candidate.

## Objective
Remove the bogus maximum task dependency-depth guard. The graph model must permit chains/subgraphs as deep as users need. A visualization or prompt concern must never reject valid graph structure. When historical depth becomes visually noisy, users can archive completed/old tasks and inspect the remaining induced subgraph(s).

## Required behavior
- Remove `guardrails.max_task_depth` as an enforced task-creation constraint from add/edit/insert/rescue/import/federation and all other graph mutation paths. There is no semantic depth maximum.
- Remove the depth ceiling from agent prompts and user-facing setup/config/TUI settings. Migrate existing config safely: accept and ignore the obsolete key with a clear one-time/read-only lint notice or migration, without breaking old projects.
- Keep cycle validation and resource safety, but express them as bounded/iterative algorithms, total-size/work budgets, cancellation, and asynchronous visualization—not as a limit on valid graph depth.
- Ensure graph algorithms that can encounter deep chains are iterative or otherwise stack-safe: load/save, readiness, critical path, dependency closure, placement, archive, reset, why-blocked, impact, federation, HTML/viz derivation, and TUI navigation.
- Visualization must handle arbitrarily deep graphs through scroll/virtualization, collapse/filter, and asynchronous bounded derivation. It may simplify presentation under load but may not hide/reject legitimate active tasks.
- Archiving old/completed tasks must yield a clear view of the remaining induced subgraph or multiple resulting connected components. Preserve dependency/history integrity and provide honest boundary markers where archived predecessors/successors were removed from the active view; never rewrite history merely to make layout shallow.
- Existing archived-task recovery, search, exact IDs, task logs, retries/evaluation, and graph commands remain correct.

## Validation
- [ ] Construct and mutate visible chains substantially deeper than 8 (including at least 1,000 nodes) without configuration overrides, rejection, recursion overflow, or quadratic UI blocking.
- [ ] Deep-chain readiness, critical path, why-blocked, impact, reset/archive, serialization/reload, service dispatch, and TUI/HTML derivation are correct and bounded; inserting a deep active task updates a running TUI promptly rather than taking tens of seconds or remaining offscreen without a discoverable pulse/jump.
- [ ] Archive an old middle/prefix region and show the remaining induced component(s) clearly with honest archived-boundary context; unarchive/recovery preserves exact edges/history.
- [ ] Cycles and malformed graphs still fail or render safely for their own reasons, never because of depth.
- [ ] Legacy `max_task_depth` config migrates/ignores safely and disappears from prompts, config UX, and Settings.
- [ ] Run fmt, clippy, full tests and large/deep graph performance smokes under bounded disk/CPU; never globally install an unmerged candidate.

Depends on

1 non-public dependency hidden

Required by

Log

2026-07-25T09:13:19.829386013+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:24.870677348+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:29.864379939+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:34.951946750+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:39.939005755+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:45.445530236+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:50.434687677+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:13:57.124724299+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:02.127032572+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:07.257247283+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:12.246527083+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:17.729078516+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:24.665681897+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:25.273525350+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:30.263438118+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:35.260462060+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:50.357391690+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:14:56.216048145+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:01.211939923+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:06.217776742+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:10.432395766+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:15.406539488+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:20.443807636+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:25.452336158+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:30.894483843+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:35.505561269+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:40.514283552+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:45.957458451+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:50.593624065+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:15:55.618761235+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:00.758356165+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:05.766302936+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:10.777496015+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:15.772416024+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:21.221128214+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:16:41.204012314+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-25T09:18:50.569437955+00:00 Task reset for retry from failed (attempt #1) — reason: stuck failed-pending-eval (eval-pipeline exhausted on gpt-5.6-sol); rerunning on zai glm-5.2 per current profile, retry-in-place to preserve 53-file WIP
2026-07-25T09:18:50.569531978+00:00 Minted evaluation pipeline evalp-37511bf9dec338a5e71a8509 for source attempt 2 (resume-in-place retry)
2026-07-25T09:18:50.569555708+00:00 error[WG-EVAL-PIPELINE-REARM]: source attempt 2 could not atomically rearm its evaluation pipeline: .flip-remove-graph-depth-guard is still active on the previous source attempt; .evaluate-remove-graph-depth-guard is still active on the previous source attempt
2026-07-25T09:22:37.785793203+00:00 Spawned by coordinator --executor pi --model zai:glm-5.2 --isolation required-worktree
2026-07-25T09:23:06.201312253+00:00 Retry-in-place: preserving prior 52-file WIP. Auditing state: build/test/smoke, fixing gaps, then committing + wg done.
2026-07-25T13:00:47.571638664+00:00 Task unclaimed: agent 'agent-726' (PID 1241132) process exited
2026-07-25T13:00:47.725990435+00:00 Minted evaluation pipeline evalp-06e2318c32ab248da4599f8e for source attempt 3 (coordinator dead-agent retry)
2026-07-25T13:00:47.725993505+00:00 error[WG-EVAL-PIPELINE-REARM]: source attempt 3 could not atomically rearm its evaluation pipeline: .flip-remove-graph-depth-guard is still active on the previous source attempt; .evaluate-remove-graph-depth-guard is still active on the previous source attempt
2026-07-25T13:00:51.748343131+00:00 Spawned by coordinator --executor pi --model zai:glm-5.2 --isolation required-worktree
2026-07-25T13:19:40.771689808+00:00 Continued from prior attempt (uncommitted WIP, EPIPE-crashed during cargo test). Validated prior work: build OK, cargo fmt --check clean, cargo clippy clean. Deep-graph smoke scenario (1001-node chain) PASSES end-to-end: legacy max_task_depth=2 accepted/ignored, deep mutation OK, iterative readiness/critical-path/impact/why-blocked/reset/trace/viz/html bounded, cycle detection intact, live TUI refresh+exact-jump, archive prefix with archived-boundary markers, undo restores exact edges.
2026-07-25T13:19:51.613293368+00:00 Test validation: cargo test --lib single-threaded => 2997 passed, 0 failed. Modified integration binaries (config/context_scope/remove_validation_cli/trace_function_layers/prompt_snapshots) => all pass. Full parallel run shows ONLY pre-existing flaky HOME-env race in profile::named::tests (racers in assign.rs/login.rs/setup.rs/secret.rs mutate process-wide HOME without the profile mutex — none modified by this task; confirmed pre-existing on base commit ab615900). Single-threaded proves depth-guard changes are correct.
2026-07-25T13:20:59.764337302+00:00 Committed and pushed 6adddd82 (wg/agent-689/remove-graph-depth-guard). Not globally installed (smoke uses candidate binary in isolated CARGO_TARGET_DIR).
2026-07-25T13:32:24.172154938+00:00 Merged origin/main (resolved tests/smoke/manifest.toml append-conflict: kept both deep_graph_archive_tui + predictive_build_admission_opt_in scenarios). Re-validated post-merge: build OK, fmt/clippy clean, lib tests single-threaded 2963 passed, modified integration tests pass, deep_graph_archive_tui smoke PASS. Pushed ee8711fe.
2026-07-25T13:32:32.601324842+00:00 Task marked as done; scheduled evaluator is advisory evidence only (execution is not a quality pass)