Replace removed computercontroller tools and rerun Goose recipes through parameter validation
Goose v1.47.0 now validates recipe parameter values, restores documented minijinja filters, isolates summon subagent notifications, and removes computercontroller tools automation_script, web_scrape, and cache. Upgrade, fix recipes and those tool calls, then use the new git-branch and recent-model chat UI.Why this loop
v1.47.0 is a breaking Goose release. computercontroller no longer ships automation_script, web_scrape, or cache, so skills and sessions that still call those tools will fail until rewritten. Recipe parameter values are now validated, and documented minijinja filters work again because builtins were re-enabled, so recipes that used to start with bad params or skip filters will change behavior. Streamable HTTP extensions can use pre-registered OAuth clients, OAuth token transport is forced to be secure, and the Gemini OAuth provider is deprecated. Concurrent summon subagents get isolated notifications. Chat shows an interactive git branch in the bottom bar and recently used models in the footer picker, and one-shot fast-model calls no longer pay the prompt-cache write premium. Context commands sent to context-owning providers are rejected. Upgrade, then fix tool calls and recipes before relying on the new session chrome.
Proposed actions
- Upgrade Goose to v1.47.0. Search the repo for computercontroller tool invocations named automation_script, web_scrape, or cache and rewrite each workflow to tools that still exist, because v1.47.0 removed those three from computercontroller.
- Re-run every Goose recipe with its real parameter values. v1.47.0 validates recipe parameter values and will reject invalid inputs that previously started a session. Re-check documented minijinja filters in those recipes; builtins were re-enabled so documented filters work again.
- For each streamable_http Goose extension, switch auth to a pre-registered OAuth client. Stop using the Gemini OAuth provider (deprecated in v1.47.0). If an OAuth handshake fails after upgrade, move token exchange onto secure transport; Goose now enforces it.
- In Goose chat, read the interactive git branch indicator in the bottom bar and pick from recently used models in the footer model picker before the first prompt. Keep one-shot fast-model calls as one-shot so Goose skips the prompt-cache write premium.
- Run concurrent summon subagents without serializing them to unmix toasts; v1.47.0 isolates concurrent subagent notifications. Do not send context commands to context-owning providers—Goose now rejects those.
Agent prompt
Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI costGoose session task
DevAgentRadar → Goose
You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.
Context
Assistant: Goose Proposal: Replace removed computercontroller tools and rerun Goose recipes through parameter validation Summary: Goose v1.47.0 now validates recipe parameter values, restores documented minijinja filters, isolates summon subagent notifications, and removes computercontroller tools automation_script, web_scrape, and cache. Upgrade, fix recipes and those tool calls, then use the new git-branch and recent-model chat UI. Primary source: https://github.com/aaif-goose/goose/releases/tag/v1.47.0
Why it matters
v1.47.0 is a breaking Goose release. computercontroller no longer ships automation_script, web_scrape, or cache, so skills and sessions that still call those tools will fail until rewritten. Recipe parameter values are now validated, and documented minijinja filters work again because builtins were re-enabled, so recipes that used to start with bad params or skip filters will change behavior. Streamable HTTP extensions can use pre-registered OAuth clients, OAuth token transport is forced to be secure, and the Gemini OAuth provider is deprecated. Concurrent summon subagents get isolated notifications. Chat shows an interactive git branch in the bottom bar and recently used models in the footer picker, and one-shot fast-model calls no longer pay the prompt-cache write premium. Context commands sent to context-owning providers are rejected. Upgrade, then fix tool calls and recipes before relying on the new session chrome.
Suggested actions
- Upgrade Goose to v1.47.0. Search the repo for computercontroller tool invocations named automation_script, web_scrape, or cache and rewrite each workflow to tools that still exist, because v1.47.0 removed those three from computercontroller.
- Re-run every Goose recipe with its real parameter values. v1.47.0 validates recipe parameter values and will reject invalid inputs that previously started a session. Re-check documented minijinja filters in those recipes; builtins were re-enabled so documented filters work again.
- For each streamable_http Goose extension, switch auth to a pre-registered OAuth client. Stop using the Gemini OAuth provider (deprecated in v1.47.0). If an OAuth handshake fails after upgrade, move token exchange onto secure transport; Goose now enforces it.
- In Goose chat, read the interactive git branch indicator in the bottom bar and pick from recently used models in the footer model picker before the first prompt. Keep one-shot fast-model calls as one-shot so Goose skips the prompt-cache write premium.
- Run concurrent summon subagents without serializing them to unmix toasts; v1.47.0 isolates concurrent subagent notifications. Do not send context commands to context-owning providers—Goose now rejects those.
Config surfaces this release may change
- skills (high confidence) — check your repo before applying
- subagent definitions — check your repo before applying
After you finish
Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.
Tell the human: open https://devagentradar.com/proposals/goose-v1-47-0-replace-removed-computercontroller-tools-and-rerun-goose-r and mark Applied, Skipped, or Failed. Proposal id: 2e96c53e-2721-4e16-a2dd-b99e9b208889
Your job
- Restate the change in one sentence.
- Propose a minimal plan for my repo (or a throwaway pilot).
- Implement only what I approve; prefer small diffs and tests.
- Call out risks (permissions, breaking APIs, cost).
Start by confirming you understood the proposal.
Your loop
This browser · no sign-in · not shared as “you”After you run the prompt
Only you can mark this. Agents cannot write your loop.
Your decision stays on this device. A public tally appears after a few votes.