Feat/summon subagent instructions by DOsinga · Pull Request #9325 · aaif-goose/goose

DOsinga · 2026-05-19T16:31:31Z

Summary

Make the summon extension handle the @agent stuff. will also call now if you just mention the name or if it just seems fitting. So "hey @pirate, where is the treasure" should work now.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 82c1740f47

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-19T16:35:05Z

+        let (tools, toolshim_tools, system_prompt) = self
            .prepare_tools_and_prompt(session_id, working_dir)
            .await?;


Keep @mention routing when tool calls are disabled

Removing resolve_at_mention from reply-context setup means @agent now relies entirely on the model calling delegate, but in Goose chat mode all non-frontend tool calls are skipped (GooseMode::Chat path in agent.rs), so explicit @name requests can no longer activate the target subagent instructions. Before this commit, @name injected matching source content directly into the system prompt and worked without tool execution; with this change, the same input can silently degrade to a normal response in chat-mode sessions.

Useful? React with 👍 / 👎.

pirate · 2026-05-19T16:40:45Z

Arrr matey, it worked, you have summoned me from a great slumber.
The treasure be buried in a far off land marked by the big red 🏴‍☠️
Good luck on ye journey.

jamadeo

nice.

As for the extension itself, do you think it should continue to be

called "summon"
(en|dis)abled from the extension page?

it is not that clear to me that toggling a box called "summon" under "extensions" would affect goose's ability to use agents I've defined

jamadeo · 2026-05-19T16:50:38Z

+        out.push_str(&format!(
+            "\n• {} — {}",
+            s.name,
+            truncate(&s.description, 200)


200 here seems a tad arbitrary? though I'm not able to find a better limit

yeah good point. also it is using a locally defined truncate instead of the utility we have. introduced two constants to do it instead

DOsinga · 2026-05-19T17:17:12Z

nice.

As for the extension itself, do you think it should continue to be

called "summon"

(en|dis)abled from the extension page?

it is not that clear to me that toggling a box called "summon" under "extensions" would affect goose's ability to use agents I've defined

I think we should probably always enable it unless there is a setting like DISSABLE_SUBAGENTS. or maybe we should have some sort of features thing?

alexhancock · 2026-05-19T17:58:24Z

it is not that clear to me that toggling a box called "summon" under "extensions" would affect goose's ability to use agents I've defined

This is my main feedback too. I think we should make one extension which does sub-agents & agents and just call it "sub-agents"

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b80d597909

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-19T20:28:29Z

+        let instructions = build_subagent_instructions(context.session.as_deref());
+
        let info = InitializeResult::new(ServerCapabilities::builder().enable_tools().build())
-            .with_server_info(Implementation::new(EXTENSION_NAME, "1.0.0").with_title("Summon"));
+            .with_server_info(Implementation::new(EXTENSION_NAME, "1.0.0").with_title("Summon"))
+            .with_instructions(instructions);


Refresh summon instructions when source files change

Build the summon instruction text lazily or refresh it per turn instead of freezing it at client construction. with_instructions(...) is populated once from a startup scan, so if a user adds/renames an agent or recipe during the same session, the model keeps an outdated name list and routing guidance and can miss explicit mentions (for example a newly created @agent) until the session/extension is recreated.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-05-19T20:28:29Z

+        return String::new();
+    };
+
+    let sources = discover_filesystem_sources(&session.working_dir);


Include recipe subrecipes in generated summon instructions

Populate instruction-time source discovery from the same source set used by runtime delegation (get_sources) or explicitly merge session.recipe.sub_recipes. As written, instruction generation only calls discover_filesystem_sources(...), which does not add SourceType::Subrecipe, so recipe-defined subrecipes are omitted from the advertised names/descriptions and the model loses the intended name/description-based routing hints for those valid delegate targets.

Useful? React with 👍 / 👎.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bb84eec559

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-19T20:54:32Z

+run_goose() {
+  local prompt="$1"
+  local outfile="$2"
+  (cd "$TESTDIR" && "$GOOSE_BIN" run --text "$prompt" --no-session 2>&1) | tee "$outfile"


Preserve goose exit status in transcript pipeline

The run_goose helper pipes Goose output through tee, but the script enables only set -e and not pipefail, so a failing Goose invocation can still return success because the pipeline status comes from tee. In failure cases (provider errors, CLI crashes, auth failures), scenarios may be evaluated against partial/error transcripts instead of stopping immediately, which makes this smoke test unreliable as a validator.

Useful? React with 👍 / 👎.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5b46d31d95

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-19T21:04:31Z

+            let mut seen: std::collections::HashSet<String> =
+                sources.iter().map(|s| s.name.clone()).collect();
+            for sr in subs {
+                if !seen.insert(sr.name.clone()) {
+                    continue;


Align instruction precedence with delegate source resolution

When a filesystem source and a recipe subrecipe share the same name, build_subagent_instructions pre-populates seen from filesystem sources and drops the subrecipe entry, but runtime resolution in get_sources gives subrecipes precedence by adding them first. This creates a mismatch where the model is shown one source description but delegate(source: "name") executes a different source, which can route work to unintended instructions for name-collision cases.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-05-19T21:04:31Z

+  else
+    echo "✗ $scenario: did NOT delegate to $source"
+    RESULTS+=("✗ $scenario")
+    return 1


Prevent errexit from stopping the smoke test after one miss

With set -e enabled, assert_delegated_to returning 1 causes the script to exit immediately when an assertion fails (e.g., in Scenario 1 or 2), so later scenarios and the final consolidated summary never run. That defeats this script’s own result-collection design and makes failures less diagnosable because you only see the first miss.

Useful? React with 👍 / 👎.

* main: (70 commits) Feat/summon subagent instructions (#9325) feat: open-plugins generalization + skills (#9112) feat(hooks): PreToolUse denial (#9304) Add support for optional api_key configuration for declarative openai-engine providers (#9202) fix(cli): use plain '> ' prompt instead of goose emoji (#9305) flag for login shell PATH (#9313) Remove popular chat topics from new chat screen (#9307) fix: stop killing goosed when a window closes (#9302) Remove vendored Windows binaries (#9318) Add Linux musl CLI builds (#9240) feat(acp): paginate session list (#9199) docs: reorganize (#9310) Structured per-provider config block, non-destructive provider switching (#8977) feat(cli): add `goose review` local code review command (#9114) feat(tui): diff viewer (#9260) fix(otel): emit trace_output as span attribute instead of event (#9255) docs: add guide for connecting goose Desktop to a remote goosed server (#9275) fix(config): check file fallback when keyring has no entry (#9279) fix(desktop): ScheduleModal error message styling (#9278) fix(ui): align sidebar hamburger in macOS fullscreen (#9257) ...

* main: (38 commits) [Prompt injection mitigation] Update pattern-based detection to reduce FPs (#9350) feat: add Harbor eval runner (#9138) chore(release): bump version to 1.35.0 (minor) (#9150) Include request URL in provider error messages (#9232) fix(databricks): ensure parallel tool image responses don't interleave tool results (#9241) Surface resolved Databricks model metadata (#9206) Add unified thinking effort control across all providers (#9242) Add Linux desktop Vulkan packages (#9323) chore: update canonical model registry (#9331) feat: slash commands (built-in, skill, recipe) in acp server (#9238) feat: add /goal command for agent self-evaluation before finishing (#9069) Feat/summon subagent instructions (#9325) feat: open-plugins generalization + skills (#9112) feat(hooks): PreToolUse denial (#9304) Add support for optional api_key configuration for declarative openai-engine providers (#9202) fix(cli): use plain '> ' prompt instead of goose emoji (#9305) flag for login shell PATH (#9313) Remove popular chat topics from new chat screen (#9307) fix: stop killing goosed when a window closes (#9302) Remove vendored Windows binaries (#9318) ...

Co-authored-by: Douwe Osinga <douwe@squareup.com>

Douwe Osinga added 3 commits May 19, 2026 11:39

A subagent by any other name

6aae57c

Simplify

772660d

Skip

82c1740

chatgpt-codex-connector Bot reviewed May 19, 2026

View reviewed changes

jamadeo approved these changes May 19, 2026

View reviewed changes

What Jack said

b80d597

chatgpt-codex-connector Bot reviewed May 19, 2026

View reviewed changes

Hmm

bb84eec

chatgpt-codex-connector Bot reviewed May 19, 2026

View reviewed changes

Remove some comments

5b46d31

chatgpt-codex-connector Bot reviewed May 19, 2026

View reviewed changes

DOsinga added this pull request to the merge queue May 19, 2026

Merged via the queue into main with commit 971c690 May 19, 2026
22 of 23 checks passed

DOsinga deleted the feat/summon-subagent-instructions branch May 19, 2026 21:52

github-actions Bot mentioned this pull request May 20, 2026

chore(release): release version 1.35.0 #9346

Closed

tushareshukla pushed a commit to tushareshukla/goose that referenced this pull request May 30, 2026

Feat/summon subagent instructions (aaif-goose#9325)

b9545d4

Co-authored-by: Douwe Osinga <douwe@squareup.com>

DOsinga mentioned this pull request Jul 31, 2026

Allow add new Agents #9394

Closed

shafqatevo pushed a commit to shafqatevo/goose that referenced this pull request Aug 7, 2026

Feat/summon subagent instructions (aaif-goose#9325)

59ae9b0

Co-authored-by: Douwe Osinga <douwe@squareup.com>

Uh oh!

Conversation

DOsinga commented May 19, 2026

Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

pirate commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jamadeo left a comment

Choose a reason for hiding this comment

Uh oh!

jamadeo May 19, 2026

Choose a reason for hiding this comment

Uh oh!

DOsinga May 19, 2026

Choose a reason for hiding this comment

Uh oh!

DOsinga commented May 19, 2026

Uh oh!

alexhancock commented May 19, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

pirate commented May 19, 2026 •

edited

Loading