feat(agent): show PostHog products used below each turn#2476
Draft
raquelmsmith wants to merge 6 commits into
Draft
feat(agent): show PostHog products used below each turn#2476raquelmsmith wants to merge 6 commits into
raquelmsmith wants to merge 6 commits into
Conversation
When the agent uses a PostHog resource via the MCP `exec` dispatcher during a turn, surface a chip row under that turn's final message listing the products touched (Experiments, Feature flags, SQL, Error tracking, …). A turn with no PostHog calls shows no row. Detection and classification live in the agent (main/business layer); the renderer is strictly UI: - packages/agent/src/posthog-products.ts: classify an exec sub-tool's domain into a stable product id (admin/meta domains hidden; unknown → generic "PostHog"). Exported via @posthog/agent. - PostToolUse hook records the product behind each executed `call` onto a per-turn Set on the session; reset at prompt start. - At turn end the agent emits a new `_posthog/resources_used` notification with the turn's products, then clears the accumulator. - Renderer: buildConversationItems handles the notification and places a `resources_used` item under the final message; ResourcesUsedView renders the chips. Tests: posthog-products classification, and buildConversationItems placement / empty-payload handling. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
Adds [resources_used]-tagged debug logging across the full chain so we can see where the per-turn PostHog products row breaks: - PostToolUse hook (hooks.ts): logs the SDK's actual tool_name for any posthog tool, whether isPostHogExecTool matched, and the extracted sub-tool — catches a tool-name/regex mismatch. - claude-agent.ts: logs each classified resource as it's recorded, and whether emitResourcesUsed fires + what it emits. - buildConversationItems.ts (renderer): logs any resource notification the builder receives and whether the RESOURCES_USED branch matched. Agent logs land in the dev logs; renderer logs in chromium.log. Temporary diagnostic — to be reverted once the root cause is found. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
A turn that counts feature flags via `SELECT count() FROM feature_flags` was tagged only "SQL" — the classifier saw the execute-sql sub-tool and had no visibility into what the query was about. The chip reflected the mechanism, not the product the user asked about. Now, for execute-sql, the raw command (which embeds the HogQL) is threaded through the PostToolUse hook to a new classifier that extracts FROM/JOIN tables and maps them to products (feature_flags -> Feature flags, experiments -> Experiments, events/persons -> Product analytics, session_replay_events -> Session replay, surveys, logs). A query touching no mapped table falls back to the generic "SQL" chip, so nothing vanishes. - posthog-products.ts: TABLE_PRODUCT map, extractSqlTables, classifyPostHogSqlQuery, classifyPostHogExecCall (returns 0..n products; SQL can touch several tables). - hooks.ts / options.ts: onPostHogResourceUsed now also receives the raw command. - claude-agent.ts: accumulates all products from classifyPostHogExecCall. - Exported the new helpers; added classifyPostHogSqlQuery / classifyPostHogExecCall tests. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
SQL table → product attribution is an exact-name match, but taking the last dotted segment meant a schema-qualified warehouse table like `stripe.feature_flags` would wrongly map to Feature flags. Now a qualified reference only maps when its schema is a known PostHog schema (`system`); any other prefix (a warehouse source) is left unmapped. Exact matching also means similarly-named tables such as `statsig_feature_flags` or `feature_flags_archive` never match a product — they fall back to the generic SQL chip. Added tests covering both cases. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
…omposer Reworks the resources-used display from a per-turn inline row into a persistent bar above the chat composer that fills in live as resources are used and shows each product only once for the whole session. Agent: - Session accumulator renamed turnResources -> sessionResources; it now lives for the whole session and is never reset between turns. - Detection emits `_posthog/resources_used` immediately for each newly-seen product (session-wide dedup), instead of buffering one emit at turn end. Removed the two turn-end emit sites and the per-prompt reset. Renderer: - New SessionResourcesBar (mirrors PlanStatusBar) rendered above the composer in SessionView. accumulateSessionResources derives the deduped, first-seen-ordered product list from the session's events, so it works for both live streaming and log replay. - Removed the inline conversation-item path: buildConversationItems no longer emits a resources_used item, the SessionUpdateView case/type are gone, and ResourcesUsedView is deleted. Tests updated/added: accumulateSessionResources (dedup, ordering, empty), buildConversationItems no longer renders an inline item, agent session-field rename. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
…rder The resources bar sat in a full-width `bg-gray-2` band with a top border, which read as a heavy separate strip. Remove both so it inherits the chat area's background and sits flush above the composer. Generated-By: PostHog Code Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
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
1 participant
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.
When the agent uses a PostHog resource via the MCP
execdispatcher during a turn, surface a chip row under that turn's final message listing the products touched (Experiments, Feature flags, SQL, Error tracking, …). A turn with no PostHog calls shows no row.Detection and classification live in the agent (main/business layer); the renderer is strictly UI:
callonto a per-turn Set on the session; reset at prompt start._posthog/resources_usednotification with the turn's products, then clears the accumulator.resources_useditem under the final message; ResourcesUsedView renders the chips.Tests: posthog-products classification, and buildConversationItems placement / empty-payload handling.
Generated-By: PostHog Code
Task-Id: f2ff4d75-f51e-4618-9e64-68ca4be237fa
Problem
Changes
How did you test this?
Automatic notifications