Remove drift detection system and port dashboard to Phoenix LiveView#24
Merged
Remove drift detection system and port dashboard to Phoenix LiveView#24
Conversation
Scaffolds a Phoenix LiveView app under packages/phoenix-dashboard/ that reads the existing .htmlgraph/htmlgraph.db SQLite database and renders a real-time activity feed with: - Multi-level nesting: Session > UserQuery > Tool Events > Subagent Events (4 levels) - Color-coded badges for tool types, models, subagents, errors, features - Descending chronological order at every level - Live updates via GenServer polling (1s) + PubSub broadcast to LiveView - Expand/collapse per conversation turn and per nested event - Tree connectors for visual hierarchy - New event flash animation Key modules: - Activity: Hierarchical query builder against agent_events table - EventPoller: GenServer that polls SQLite and broadcasts via PubSub - Repo: Lightweight read-only SQLite access via exqlite - ActivityFeedLive: LiveView with recursive event_row component https://claude.ai/code/session_01R5FpDt5NCEkgVdXJSfddQ4
Force-add Elixir source files that were excluded by root .gitignore lib/ rule. Includes all application modules: - Activity: hierarchical event query builder - Repo: read-only SQLite access via exqlite - EventPoller: GenServer polling + PubSub broadcast - ActivityFeedLive: LiveView with recursive nested events - Styles: inline CSS (dark theme, badges, tree connectors) - Layouts, Router, Endpoint, CoreComponents https://claude.ai/code/session_01R5FpDt5NCEkgVdXJSfddQ4
- Update Phoenix 1.8 + LiveView 1.1 + exqlite API compatibility - Fix CSS loading (HEEx doesn't interpolate inside <style> tags) - Move inline CSS to static asset pipeline (Tailwind + esbuild) - Add multi-level event hierarchy (Session → UserQuery → Tools → Task) - Remove UserQuery badge (implicit in hierarchy structure) - Sanitize summaries: strip XML tags, raw JSON metadata dumps - Filter empty PreToolUse events (dedup with PostToolUse) - Add color-coded tool chips (Bash=green, Read=blue, Edit=yellow, etc.) - Add subagent re-parenting and orphan event adoption - Configure esbuild/tailwind asset watchers for dev mode Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add phx-key attributes on session groups, turns, and events for stable DOM identity (eliminates flicker) - Debounce event reload to 500ms (prevents rapid re-renders) - Remove new_event_ids highlight mechanism (caused double re-render) - Session cards now show last prompt text instead of raw UUIDs - Derive session status from data: active/idle/completed (idle = no events in 30 min) - Remove NULL AS model — use real model column from sessions table - Suppress "system" agent label when not meaningful - Add badge-status-idle CSS (amber) and session-subtitle styling Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Filter noise events (Stop, SessionResume, InstructionsLoaded) in has_meaningful_content instead of letting them leak through - Move compute_stats to run on displayed_children (after filter) so tool count badge matches what's actually shown - Orchestrator Bash/MCP/Playwright calls now visible alongside Task delegation events in expanded turns Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
All child queries and merge function now use DESC ordering, consistent with the UserQuery turn ordering. Interleaved Task delegations and orchestrator tool calls are now chronologically clear with most recent events at the top. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Only filter known noise events (Stop, SessionResume, InstructionsLoaded, SessionStart, SessionEnd). Orchestrator tool calls with JSON metadata summaries that sanitize to empty are now kept — they display as tool chip + timestamp, which correctly represents the work performed. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove (htmlgraph:haiku-coder): prefix from Task event summaries since the agent type is already displayed as a badge. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add step_id field to Step model (format: step-{feature_id}-{index})
- Emit data-step-id in HTML, parse it back in parser.py
- Auto-generate step_ids on feature creation and builder save
- Add step_id column to agent_events table with migration + index
- Add resolve_active_step() to event_tracker — finds first incomplete step
- Pass step_id through record_event_to_sqlite() for every event
- 34 new tests covering model, parser, DB, resolution, backward compat
- All 2213 existing tests pass (0 failures)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add RelationshipType enum (blocks, relates_to, spawned_from, etc.) - Add Node convenience methods: relates_to(), spawned_from(), caused_by(), implements() - Add Builder methods: .relates_to(), .spawned_from(), .caused_by(), .implements() - Add Collection edge methods: add_edge() with dual-write (HTML + SQLite), related_to(), blocked_by(), blocks(), edges_of() queries - Add DB methods: insert_graph_edge(), get_graph_edges(), delete_graph_edge() - Add edge_sync module: sync_html_edges_to_sqlite(), sync_sqlite_edge_to_html() - 42 new tests, all 2255 tests pass (0 failures) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 3a — Critical path analysis: - New analytics/critical_path.py with find_critical_path(), find_bottlenecks(), get_dependency_graph() - Topological sort on graph_edges, cycle detection, DAG visualization Phase 3b — Step dependencies: - Step.depends_on field for intra-feature step ordering - Node.get_ready_steps() returns steps with met dependencies - next_step property respects dependency ordering - Parser extracts data-depends-on from HTML Phase 3c — Auto-divergence detection: - FeatureWorkflow.auto_create_divergent_feature() with spawned_from edge - CIGS guidance now shows active step context - Divergence scoring on Task events (keyword overlap heuristic) 34 new tests (17 auto-divergence + 17 critical path), all passing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: aadf939b3e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+15
to
+17
| if Path.type(path) == :relative do | ||
| Path.join(Application.app_dir(:htmlgraph_dashboard, ""), path) | ||
| |> Path.expand() |
There was a problem hiding this comment.
When HTMLGRAPH_DB_PATH is unset, db_path/0 joins "../../.htmlgraph/htmlgraph.db" against Application.app_dir/2, which points to _build/.../lib/htmlgraph_dashboard in Mix projects. That resolves to _build/dev/.htmlgraph/htmlgraph.db instead of the repository’s .htmlgraph/htmlgraph.db, so all queries fail to open the DB and the live feed appears empty by default.
Useful? React with 👍 / 👎.
Comment on lines
+211
to
+214
| flat | ||
| |> Enum.map(fn c -> (c["execution_duration_seconds"] || 0) end) | ||
| |> Enum.sum() | ||
| |> Float.round(2), |
There was a problem hiding this comment.
compute_stats/1 pipes Enum.sum() into Float.round/2; for a UserQuery with no child events, flat is empty and Enum.sum([]) returns integer 0, which raises FunctionClauseError in Float.round/2. This crashes feed construction for sessions that include turns without tool activity.
Useful? React with 👍 / 👎.
…ok-htmlgraph-SKUV1 # Conflicts: # .claude-plugin/marketplace.json # packages/claude-plugin/.claude-plugin/marketplace.json # packages/claude-plugin/.claude-plugin/plugin.json # packages/gemini-extension/gemini-extension.json # pyproject.toml # src/python/htmlgraph/__init__.py # src/python/htmlgraph/builders/base.py # src/python/htmlgraph/hooks/prompt_analyzer.py
shakestzd
force-pushed
the
claude/explore-livebook-htmlgraph-SKUV1
branch
from
aadf939 to
3715aa5
Compare
…gent updates Recover files from the 11 commits (2e7cca1..aadf939) that were overwritten during force-push: - subnode.py: CSS selector query/update layer for HTML elements - test_subnode.py: test suite for sub-node operations - check-module-size.py: module size enforcement script - MODULE_REFACTORING_TRACK.md: refactoring plan for 15 oversized modules - Agent definitions: updated with module size standards - code-hygiene.md: module size & complexity standards - user-prompt-submit.py: updated hook script - system-prompt-default.md: updated system prompt Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Deploy v0.33.79 (Phase 3: auto-divergence, step deps, critical path) - Create track trk-92f36021: Phoenix Dashboard SDK Integration - Plan: replace raw SQLite with Python SDK via Elixir Port bridge Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR removes the deprecated drift detection and auto-classification system from HtmlGraph hooks, and introduces an exploratory Phoenix LiveView dashboard to replace the static HTML activity feed with real-time WebSocket-driven updates.
Key Changes
Drift System Removal
src/python/htmlgraph/hooks/drift_handler.pyandsrc/python/htmlgraph/hooks/drift.py— centralized drift detection logic no longer neededDriftQueueManagerclass fromstate_manager.py— drift queue persistence eliminateddrift_scorefield from activity tracking across:session_manager.py— no longer calculates per-activity drift scoressessions/activity.py— removed drift calculation from activity trackingsessions/features.py— removed drift_score from attribution resultsmodels.pyandmodels/context.py— removed drift_score from ActivityEntry schematest_drift_queue_cleanup.pyand drift-related test cases fromtest_state_manager.pyandtest_event_tracker.pyPhoenix LiveView Dashboard (Exploratory)
packages/phoenix-dashboard/— complete Phoenix/Elixir applicationactivity.ex— queries HtmlGraph SQLite database and builds multi-level nested activity tree (Session → UserQuery → Tool Events → Subagent Events)event_poller.ex— GenServer that polls SQLite every 1 second and broadcasts new events via Phoenix PubSubrepo.ex— read-only SQLite3 adapter using exqliteactivity_feed_live.ex— LiveView component with expand/collapse state, real-time updates, and agent filteringstyles.ex— inline CSS module with GitHub-inspired dark theme (no build tools required)Code Standards Enforcement
scripts/check-module-size.py— enforces module size limits (warn >300 lines, fail >500 for new modules, critical >1000)tests/test_subnode.py— tests for sub-node HTML element querying and mutation layersrc/python/htmlgraph/subnode.py— CSS selector-based querying and updating of HTML elements within track/feature filesdocs/tracks/MODULE_REFACTORING_TRACK.md— comprehensive refactoring plan addressing 15 oversized modules and code duplication.pre-commit-config.yamlSupporting Changes
focus_ratiofield toSessionclass to measure activity attribution coverage (replaces per-event drift_score)test_drift_parent_activity.pyto test session-level focus_ratio instead of per-event drift scores.htmlgraph/htmlgraph.dbschema (binary change)Implementation Notes
.htmlgraph/htmlgraph.dbSQLite file with no schema changes requiredhttps://claude.ai/code/session_01R5FpDt5NCEkgVdXJSfddQ4