Add ambient LLM incident watcher with AI agents#321
Merged
Conversation
Phase 1 of #305: introduces the watcher pane layout with a bordered viewport below the table, toggle via 'w' key, dynamic table height adjustment, resize handling, and status header showing provider name, health, and activity state. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Phase 2 of #305: redirect /agent CLI responses from the incident viewer to the watcher pane. Add ring buffer for observation history, configurable emoji/text markers (🤖/📡 vs ☻/☺) via emoji config bool, per-line marker prefixing, word wrap at viewport width, mouse wheel scrolling, and persistent input field during agent queries. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Mirrors vim behavior: pressing / opens command input with the slash already typed, ready for /agent or /flag commands. The i and : keys continue to open input without pre-fill. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Use vim-style : prefix for commands (:agent, :flag, :flags, :unflag). Reserve / for future search. Remove f and i key bindings — use : or / to enter input mode. Both : and / pre-fill their character in the input field. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Phase 4 of #305: pure Go detectors that analyze the incident list for cross-incident patterns — service storms (3+ on same service), cluster storms (2+ on same cluster), and high urgency clustering. Detectors run on incident list updates and flag changes. Observations are deduplicated with a 5-minute cooldown. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Set viper.SetDefault("emoji", true) so emoji markers are used when
the config key is absent. Auto-expand the watcher pane when detectors
produce new observations.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Phase 5 of #305: when a pattern detector fires and the LLM provider is healthy, dispatch a Query to synthesize a richer natural-language observation. Falls back to raw heuristic text when the provider is unavailable or offline. Watcher header shows analyzing state. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Flag changes (add, remove, clear, load) rebuilt the match cache but never triggered a table row rebuild — markers only appeared on the next 15-second poll. Now all flag operations dispatch an immediate table rebuild. Flags load also kicks off OCM enrichment for any unenriched incidents so cluster-based flags match as soon as enrichment completes. Also rebuild flag cache when cluster enrichment arrives so new cluster data is matched against existing flags. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Type :watcher <question> to query the configured LLM provider with rich incident context: selected incident details, alerts, cluster info, service logs, limited support history, notes, and the full incident queue summary. Responses appear in the watcher pane with the watcher marker prefix. Input stays focused for follow-up queries. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
buildWatcherContext now reads alerts and notes from incidentCache (populated by OCM enrichment) instead of selectedIncidentAlerts/Notes (only populated on manual view). Also includes SOP links, cloud provider, and extracts cluster context into a reusable helper. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Table always gets at least 10 rows. When watcher is expanded: if enough room, split 2/3 table and 1/3 watcher (default 10 rows each minimum). If tight, table gets its minimum and watcher gets the rest (min 5 rows). All layout constants are named, no magic numbers. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Move the [AI Watcher] status from its own header line into the existing footer as a right-justified column alongside "Watching for updates...". Only shown when the watcher pane is expanded. Saves one line of vertical space for the table and watcher content. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Convert all error paths in agent CLI, watcher query, and provider status checks from permanent setStatus to flashNotification so errors appear temporarily in the status bar and auto-clear. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Pad idle state to match the width of spinner + analyzing label so the footer line doesn't wrap when transitioning between states. Bold the analyzing text when active. Use lipgloss.Width for spinner measurement instead of a magic number. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Revert persistent input focus — all commands (:agent, :watcher, :flag) now blur the input and return focus to the table immediately after dispatch, matching vim's :w behavior. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Use theme highlight color (white) instead of bold for the analyzing label. Add MaxWidth to both footer columns to prevent line wrapping. Remove idle padding — MaxWidth handles overflow by truncation. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Render the watcher status first, measure its actual width including ANSI codes via lipgloss.Width, then give the left column the remaining space. Eliminates wrapping regardless of spinner or styled text content. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Responses from :agent, :watcher, and LLM synthesis now display word-by-word in the watcher pane at 3 words per 30ms tick, giving a fast typewriter effect. Uses tea.Tick and a SetLast buffer method to update the last entry in-place as words are revealed. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Show a counting-up timer next to "analyzing..." in the footer status when an agent or watcher query is in flight. The spinner already triggers re-renders so the timer updates live. Resets when a new query starts. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Count down from the query timeout (60s for agent/watcher, 30s for synthesis) instead of counting up. Extract timeout constants from hardcoded values into named constants. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Add tea.MouseMsg to the high-frequency message filter to prevent mouse motion events from flooding the journal. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Add ocmServiceLogsMsg and limitedSupportMsg to the high-frequency message filter to prevent large OCM responses from flooding the journal. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
The agent CLI now receives the same rich context as the watcher: selected incident details, alerts, cluster info, service logs, limited support, notes, and full queue summary — all piped via stdin alongside the system prompt and user query. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Add Padding(0, 1) to the watcher container style to match the table cell padding, aligning text indentation between the two panes. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Move agent and watcher system prompts from hardcoded constants to configurable agent_system_prompt and watcher_system_prompt config keys with sensible defaults. Both :agent and :watcher now use the same buildWatcherContext for incident data. Extract cell padding into named constants for consistency between table and watcher pane. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Replace strings.Fields (which collapses all whitespace) with a custom splitter that preserves newlines as separate tokens. The typewriter now reconstructs multi-line responses correctly. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
…cher Phase 6 of #305: - New docs/ai-agents.md: quickstart, architecture, commands, context, prompts, markers, health checks, detectors, timeouts, privacy - New docs/configuration.md: CLI args section + full config file reference with every key, type, default, and description - Update docs/flag-conditions.md: /flag -> :flag prefix, remove f key - Update docs/llm-providers.md: add watcher integration section - Update README.md: add AI agents and config doc features, update key bindings (remove f/i, add w), update command prefixes, link to new docs, add emoji/prompt config keys to optional table - New docs/plans/062-ambient-watcher.md: plan doc for the branch Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Fix gofmt alignment in keymap.go, model.go, watcher.go. Add tests for SetLast, buildIncidentSummary, isWatcherCommand, parseWatcherQuery, splitKeepingNewlines, and buildWatcherContext. Escape pipe character in configuration.md markdown tables. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #321 +/- ##
==========================================
+ Coverage 67.75% 68.30% +0.55%
==========================================
Files 45 46 +1
Lines 6906 7323 +417
==========================================
+ Hits 4679 5002 +323
- Misses 1915 1988 +73
- Partials 312 333 +21 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
…arker The flag_marker config key is effectively dead — markers are controlled by the emoji toggle. Update documentation to point users to emoji: false instead of flag_marker config. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
30+ tests covering: watcher toggle, prompt/response/synthesis message handlers, typewriter ticks, mouse events, input mode :watcher dispatch, renderWatcherPane, renderWatcherStatus, renderFooter watcher branch, buildClusterContext, advanceTypewriter, and buildWatcherContext with cached alerts and notes. Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Use tea.MouseMsg{Action: MouseActionPress, Button: MouseButtonWheelDown}
instead of the deprecated tea.MouseMsg{Type: MouseWheelDown}.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
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
:agentcommand (CLI subprocess) and:watchercommand (LLM API) with shared rich incident context (alerts, OCM cluster info, service logs, limited support, notes)agent_system_prompt,watcher_system_prompt), emoji toggle, andagent_cli_command/to:(vim-style), add/key for future search, removefandikey bindingsCloses #305
Closes #312
Closes #315
Test plan
make test-allpasses (fmt-check, vet, lint, test, test-race)w, resizes dynamically, scrolls with mouse wheel:agentqueries dispatch to CLI agent with full incident context, responses typewrite into watcher pane:watcherqueries dispatch to LLM provider with full context, responses typewrite into watcher pane:flag/:flags/:unflagcommands work with new:prefixemoji: falseswitches all markers to text fallbackssrepd --dev) works with watcher features🤖 Generated with Claude Code