Home

mweinbach / agent-coworker

publicmweinbach/agent-coworker
Overview Code History Branches Pull requestsIssuesInsights
main
HomeOverview Code PRsIssues

Migrate model prompts to Markdown and consolidate the built-in toolbelt (#142)

4 months ago

2409dab
Authored
Max Weinbach6/1/2026, 4:59:14 AM
* feat: refresh model prompt metadata

* fix: remove model-visible usage tool

* feat: migrate model prompts to markdown

* refactor: remove notebookEdit tool and rename ask to AskUserQuestion

Remove the notebookEdit tool end-to-end (impl, registration, codex
boundary, child-agent roles, write-path permissions, all 12 model
prompts, tests, docs, and harness eval scenarios). The repo has no
notebooks and the read tool has no cell-aware support to target one.

Consolidate the clarifying-question tool to a single, clearer
model-facing name:
- Drop the redundant AskUserQuestion alias that registered the same
  tool under two names.
- Rename `ask` -> `AskUserQuestion` across the registration, all 12
  model prompts (declarations + prose), harness, tests, and docs.

The internal `type: "ask"` interaction event, createAskTool factory,
and ctx.askUser plumbing are intentionally unchanged.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix: address model prompt review findings

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Parent3bf5455

83 files changed
  • GEMINI.md+1−1
  • README.md+2−4
  • apps/mobile/src/cowork-shared/googleThinking.ts+1−0
  • config/models/anthropic/claude-haiku-4-5.json+1−1
  • config/models/anthropic/claude-opus-4-6.json+1−1
  • config/models/anthropic/claude-opus-4-7.json+3−3
  • config/models/anthropic/claude-opus-4-8.json+1−1
  • config/models/antigravity/gemini-3.5-flash.json
+1
−1
  • config/models/codex-cli/gpt-5.3-codex-spark.json+1−1
  • config/models/codex-cli/gpt-5.4-mini.json+1−1
  • config/models/codex-cli/gpt-5.4.json+1−1
  • config/models/codex-cli/gpt-5.5.json+1−1
  • config/models/google/{gemini-3.1-flash-lite-preview.json => gemini-3.1-flash-lite.json}+2−2
  • config/models/google/gemini-3.5-flash.json+1−1
  • config/models/openai/gpt-5.4-mini.json+1−1
  • config/models/openai/gpt-5.4-nano.json+14−0
  • config/models/openai/gpt-5.4-pro.json+14−0
  • config/models/openai/gpt-5.4.json+1−1
  • config/models/openai/gpt-5.5.json+1−1
  • docs/architecture.md+2−4
  • packages/harness/src/run_raw_agent_loops.ts+13−15
  • prompts/system-models/claude-haiku-4-5.json+0−21
  • prompts/system-models/claude-haiku-4-5.md+557−0
  • prompts/system-models/claude-opus-4-6.json+0−9
  • prompts/system-models/claude-opus-4-6.md+541−0
  • prompts/system-models/claude-opus-4-7.json+0−9
  • prompts/system-models/claude-opus-4-7.md+543−0
  • prompts/system-models/claude-opus-4-8.json+0−9
  • prompts/system-models/claude-opus-4-8.md+543−0
  • prompts/system-models/claude-sonnet-4-6.md+15−16
  • prompts/system-models/gemini-3-flash-preview.md+17−18
  • prompts/system-models/gemini-3.1-pro-preview.md+17−19
  • prompts/system-models/gemini-3.5-flash.md+423−0
  • prompts/system-models/gpt-5.2.md+16−21
  • prompts/system-models/gpt-5.4.json+0−21
  • prompts/system-models/gpt-5.4.md+717−0
  • prompts/system-models/gpt-5.5.md+727−0
  • prompts/system.md+21−28
  • src/models/childAgentModelInfo.ts+3−1
  • src/models/registry.ts+7−2
  • src/prompt.ts+26−133
  • src/runtime/codexAppServer/config.ts+1−1
  • src/runtime/googleInteractionsModel.ts+8−12
  • src/runtime/openaiNativeResponses.ts+21−3
  • src/runtime/openaiResponsesModel.ts+3−1
  • src/server/agents/roles.ts+0−4
  • src/server/title/remoteModelTitle.ts+1−1
  • src/session/pricing.ts+17−0
  • src/shared/googleThinking.ts+1−0
  • src/tools/codexBoundary.ts+3−2
  • src/tools/index.ts+0−8
  • src/tools/notebookEdit.ts+0−111
  • src/tools/usage.ts+0−205
  • src/utils/permissions.ts+1−1
  • test/agent.test.ts+2−5
  • test/harness.ws.e2e.test.ts+0−1
  • test/models.registry.test.ts+53−0
  • test/prompt.test.ts+118−63
  • test/providers/google.test.ts+16−1
  • test/providers/live-api.integration.test.ts+5−2
  • test/providers/openai.test.ts+11−0
  • test/repl.test.ts+1−3
  • test/runtime.openai-responses-runtime.test.ts+33−1
  • test/runtime.pi-runtime.test.ts+80−1
  • test/session-title-service.test.ts+4−4
  • test/session/pricing.test.ts+27−1
  • test/shared/googleThinking.test.ts+18−3
  • test/tools.usage.test.ts+0−151
  • test/tools/tools.ask.test.ts+0−1
  • test/tools/tools.bash.test.ts+0−1
  • test/tools/tools.createTools.test.ts+4−10
  • test/tools/tools.edit.test.ts+0−1
  • test/tools/tools.glob.test.ts+0−1
  • test/tools/tools.grep.test.ts+0−1
  • test/tools/tools.harness.ts+0−2
  • test/tools/tools.memory.test.ts+0−1
  • test/tools/tools.notebookEdit.test.ts+0−335
  • test/tools/tools.read.test.ts+0−1
  • test/tools/tools.skill.test.ts+0−1
  • test/tools/tools.todoWrite.test.ts+0−5
  • test/tools/tools.webFetch.test.ts+0−1
  • test/tools/tools.webSearch.test.ts+0−1
  • test/tools/tools.write.test.ts+0−1