4 Commits

Author SHA1 Message Date
steve 0acaa8c9a5 run: guard empty text part in runAgent + drop cross-repo doc ref (gadfly #16)
executus CI / test (pull_request) Successful in 1m46s
Every reviewer flagged that runAgent appended llm.Text(input) unconditionally, so
an image-only run (blank prompt) emitted an empty TextPart — inconsistent with the
sibling runSession.AttachImages which guards it. Mirror that guard
(strings.TrimSpace(input) != ""). Also:
- copy opts before appending (variadic backing array can have spare capacity; avoid
  aliasing a caller's slice).
- reword the doc comment to drop the mort-agentexec reference (executus is a
  standalone lib; a consumer name doesn't belong in its godoc).

Tests: image+text are co-located in ONE user message; an image-only run emits no
blank TextPart.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-28 01:11:15 -04:00
steve a35c176b42 run: fold inv.Images into the initial user message (multimodal opening turn)
executus CI / test (pull_request) Successful in 46s
Adversarial Review (Gadfly) / review (pull_request) Successful in 6m5s
The executor passed only the text `input` to majordomo's agent.Run, silently
dropping inv.Images — so a multimodal run (vision: chatbot @mention, chat API)
lost its images on the executus path. majordomo's Run input arg is text-only, so
fold the images into the first user message (text + image parts) via WithHistory
and call Run with empty input, mirroring mort agentexec's multimodal seeding. The
image-less path is unchanged (prompt passes straight through).

Tests: a run with Images carries the image bytes + prompt into the first model
request; the text-only path still reaches the model.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-28 00:37:53 -04:00
steve 1cf46c9954 ci: track gadfly's v1 release tag instead of a pinned sha (#15)
executus CI / test (push) Successful in 52s
2026-06-28 04:08:30 +00:00
steve 56baac758d ci: inherit gadfly's default swarm (slim caller, re-pin @b02b11d) (#14)
executus CI / test (push) Successful in 50s
2026-06-28 02:48:25 +00:00
3 changed files with 152 additions and 4 deletions
+4 -3
View File
@@ -38,9 +38,10 @@ jobs:
&& (github.actor == 'steve'
|| github.actor == 'fizi'
|| github.actor == 'dazed'))
# Pinned to an immutable gadfly commit (not @main): a push to gadfly can't
# silently change the code that runs with our forwarded secrets.
uses: steve/gadfly/.gitea/workflows/review-reusable.yml@b02b11d69139843665da4cdbf776bc0b3583490d
# Tracks gadfly's v1 release tag — a curated pointer re-moved on each release
# (unlike @main, which moves on every push). Central swarm tuning propagates
# here automatically; the tradeoff vs a full sha pin is that v1 is mutable.
uses: steve/gadfly/.gitea/workflows/review-reusable.yml@v1
# Least privilege: forward only the review secrets (not `secrets: inherit`,
# which would expose every repo secret). GITEA_TOKEN is the automatic token.
secrets:
+27 -1
View File
@@ -4,6 +4,7 @@ import (
"context"
"errors"
"fmt"
"strings"
"time"
"gitea.stevedudenhoeffer.com/steve/majordomo/agent"
@@ -320,7 +321,7 @@ func (e *Executor) Run(ctx context.Context, ra RunnableAgent, inv tool.Invocatio
// One WithSteer drains BOTH the session mailbox (a tool's AttachImages) and
// the critic's nudges before each step.
steer := func() []llm.Message { return append(mailbox.drain(), critic.drainSteer()...) }
runRes, runErr := ag.Run(runCtx, input, agent.WithSteer(steer))
runRes, runErr := runAgent(runCtx, ag, input, inv.Images, agent.WithSteer(steer))
status := statusFor(runCtx, runErr)
if runRes != nil {
@@ -440,3 +441,28 @@ func detach(ctx context.Context) context.Context {
_ = cancel // bounded by the timeout; nothing to cancel early
return c
}
// runAgent dispatches the majordomo agent loop. majordomo's Run takes a text-only
// input arg, so when the invocation carries images they're folded into the first
// user message (text + image parts) via WithHistory and Run is called with an
// empty input — the model then sees a multimodal opening turn. The image-less path
// passes the prompt straight through.
//
// The text part is omitted when input is blank (image-only run), matching
// runSession.AttachImages so no empty TextPart is sent.
func runAgent(ctx context.Context, ag *agent.Agent, input string, images []llm.ImagePart, opts ...agent.RunOption) (*agent.Result, error) {
if len(images) == 0 {
return ag.Run(ctx, input, opts...)
}
parts := make([]llm.Part, 0, len(images)+1)
if strings.TrimSpace(input) != "" {
parts = append(parts, llm.Text(input))
}
for _, img := range images {
parts = append(parts, img)
}
// Copy opts before appending so a caller-supplied backing array is never
// mutated/aliased (the variadic slice can have spare capacity).
opts = append(opts[:len(opts):len(opts)], agent.WithHistory([]llm.Message{llm.UserParts(parts...)}))
return ag.Run(ctx, "", opts...)
}
+121
View File
@@ -0,0 +1,121 @@
package run_test
import (
"context"
"strings"
"testing"
"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
"gitea.stevedudenhoeffer.com/steve/majordomo/provider/fake"
"gitea.stevedudenhoeffer.com/steve/executus/run"
"gitea.stevedudenhoeffer.com/steve/executus/tool"
)
// TestExecutorFoldsInitialImages: when the invocation carries Images, they're
// folded into the first user message (alongside the prompt text) instead of being
// dropped — majordomo's Run input arg is text-only, so the executor seeds the
// multimodal opening turn via history.
func TestExecutorFoldsInitialImages(t *testing.T) {
fp := fake.New("fake")
fp.Enqueue("m", fake.Reply("saw the image"))
m, _ := fp.Model("m")
img := llm.ImagePart{MIME: "image/png", Data: []byte("PNGDATA")}
inv := tool.Invocation{RunID: "r1", Images: []llm.ImagePart{img}}
ex := run.New(run.Config{
Registry: tool.NewRegistry(),
Models: func(ctx context.Context, _ string) (context.Context, llm.Model, error) { return ctx, m, nil },
})
res := ex.Run(context.Background(), run.RunnableAgent{ModelTier: "m"}, inv, "describe this")
if res.Err != nil {
t.Fatalf("run error: %v", res.Err)
}
calls := fp.Calls()
if len(calls) == 0 {
t.Fatal("no model calls recorded")
}
// The text + image must be CO-LOCATED in a single user message (not split
// across two), so the model reads them as one multimodal turn.
coLocated := false
for _, msg := range calls[0].Request.Messages {
sawImage, sawText := false, false
for _, p := range msg.Parts {
switch pp := p.(type) {
case llm.ImagePart:
if string(pp.Data) == "PNGDATA" {
sawImage = true
}
case llm.TextPart:
if strings.Contains(pp.Text, "describe this") {
sawText = true
}
}
}
if sawImage && sawText {
coLocated = true
}
}
if !coLocated {
t.Error("image + prompt text were not folded into the SAME user message")
}
}
// TestExecutorImageOnlyNoBlankText: an image-only run (blank prompt) must NOT emit
// an empty TextPart — the message carries just the image, matching
// runSession.AttachImages's guard.
func TestExecutorImageOnlyNoBlankText(t *testing.T) {
fp := fake.New("fake")
fp.Enqueue("m", fake.Reply("saw it"))
m, _ := fp.Model("m")
inv := tool.Invocation{RunID: "r3", Images: []llm.ImagePart{{MIME: "image/png", Data: []byte("IMG")}}}
ex := run.New(run.Config{
Registry: tool.NewRegistry(),
Models: func(ctx context.Context, _ string) (context.Context, llm.Model, error) { return ctx, m, nil },
})
res := ex.Run(context.Background(), run.RunnableAgent{ModelTier: "m"}, inv, " ")
if res.Err != nil {
t.Fatalf("run error: %v", res.Err)
}
for _, msg := range fp.Calls()[0].Request.Messages {
for _, p := range msg.Parts {
if tp, ok := p.(llm.TextPart); ok && strings.TrimSpace(tp.Text) == "" {
t.Error("image-only run emitted a blank TextPart")
}
}
}
}
// TestExecutorTextOnlyUnchanged: with no Images, the prompt flows through as the
// text input (regression guard that the fold path didn't break the common case).
func TestExecutorTextOnlyUnchanged(t *testing.T) {
fp := fake.New("fake")
fp.Enqueue("m", fake.Reply("ok"))
m, _ := fp.Model("m")
ex := run.New(run.Config{
Registry: tool.NewRegistry(),
Models: func(ctx context.Context, _ string) (context.Context, llm.Model, error) { return ctx, m, nil },
})
res := ex.Run(context.Background(), run.RunnableAgent{ModelTier: "m"}, tool.Invocation{RunID: "r2"}, "plain prompt")
if res.Err != nil {
t.Fatalf("run error: %v", res.Err)
}
calls := fp.Calls()
if len(calls) == 0 {
t.Fatal("no model calls recorded")
}
sawText := false
for _, msg := range calls[0].Request.Messages {
for _, p := range msg.Parts {
if tp, ok := p.(llm.TextPart); ok && strings.Contains(tp.Text, "plain prompt") {
sawText = true
}
}
}
if !sawText {
t.Error("text-only prompt did not reach the model")
}
}