Skip to content

[codex] clarify web search action labels#23

Merged
OmGuptaIND merged 3 commits into
mainfrom
feat/composer-context-gauge
May 4, 2026
Merged

[codex] clarify web search action labels#23
OmGuptaIND merged 3 commits into
mainfrom
feat/composer-context-gauge

Conversation

@OmGuptaIND
Copy link
Copy Markdown
Contributor

Summary

  • Show explicit search and research queries in chat action chips instead of ambiguous generic labels.
  • Normalize MCP-prefixed tool names such as anton:web_search before choosing labels, favicons, metadata, and citation source handling.
  • Preserve the existing message-list stick-to-bottom behavior while adding focus/visibility recovery.

Root Cause

Search tools can arrive at the desktop renderer with provider or MCP prefixes, so the previous exact web_search / web_research label cases were skipped and the UI fell back to generic tool names without the query target.

Impact

Users can now see what the agent searched for directly in collapsed action rows, for example Web Search "NVIDIA Vera Rubin roadmap", making web lookup activity less ambiguous.

Validation

  • git diff --check
  • pnpm --filter @anton/desktop typecheck

Note: the typecheck emits existing engine warnings for packages that want Node >=22 while this environment is on Node v20.18.1, but the desktop typecheck passed.

OmGuptaIND and others added 3 commits May 3, 2026 16:05
Lay the protocol + sizing groundwork for the composer Context gauge:
ContextBreakdown / AiContextUpdateMessage on the wire, and
SessionPromptLayerSizes + emptyPromptLayerSizes() so the Pi-SDK path
can report per-layer char budgets without re-running layer builders.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
In-composer circular gauge plus floating popover with per-category
token usage (System prompt, System tools, MCP tools, Skills, Memory
files, Messages, Autocompact buffer, Free space). Pi-SDK sessions
report the full split; harness sessions report Messages + Free space.

Wires breakdown emission on session_created, every turn_end, model
switch, connector refresh, conversation-context load, and surface
change. Calibrates char/4 estimates against the model's reported
input_tokens via a pre-turn snapshot captured at turn_start, with
the resulting scale factor persisted in PersistedSession so resumed
sessions keep their calibration. Codex harness emits per-turn
last.inputTokens (not the cumulative total) capped at the model
context window.

Replaces the old memory-count ContextIndicator badge; the
"View memory details" link in the popover footer keeps the side
panel reachable from the composer.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@OmGuptaIND OmGuptaIND marked this pull request as ready for review May 4, 2026 08:52
@OmGuptaIND OmGuptaIND merged commit c88a4b8 into main May 4, 2026
@OmGuptaIND OmGuptaIND deleted the feat/composer-context-gauge branch May 4, 2026 08:52
OmGuptaIND added a commit that referenced this pull request May 4, 2026
### Other
- chore(tasks): remove "Start a task" composer from All tasks view (#24)
- [codex] clarify web search action labels (#23)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant