cli: opencode run --format json drops the final step_finish event #53207

Activity

1. github-actions

   assigned

   jlongster

   on Oct 4, 2026
2. github-actions commented on Oct 4, 2026

   

   github-actionsBot

   on Oct 4, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be related to earlier reports. Please check:

   • run --format json can exit before emitting final step_finish event #26855: opencode run --format json completed without emitting the final step_finish event (older v1 report; same observable symptom).
   • opencode run --format json drops text and step-finish events in containerized environments #31435: opencode run --format json dropped text and step-finish events when event delivery raced with idle status (older v1 report).
   • opencode run --attach exits 0 after first tool-call step_finish, drops final text event (1.17.20 + 1.18.4) #38661: opencode run --attach exited after the initial tool-call step and dropped later output (related v1 event-loss behavior).

   These reports were for v1; this issue describes the v2 client.
3. nkootstra

   changed the title [-]`opencode run --format json` drops the final `step_finish` event[/-] [+]opencode run --format json drops the final step_finish event[/+] on Oct 5, 2026
4. nkootstra

   changed the title [-]opencode run --format json drops the final step_finish event[/-] [+]cli: opencode run --format json drops the final step_finish event[/+] on Oct 5, 2026
5. kkwalking

   added a commit that references this issue on Oct 11, 2026

   Anchor the opencode usage tests to a tool step under the v2 defect

   ...

   b046c6a
6. oldantique commented on Oct 12, 2026

   

   oldantique

   on Oct 12, 2026

   More actions

   Still reproduces on v2.0.26 (9b4ec57), Linux x86_64, deepseek/deepseek-v4-flash: 8/8 --format json runs ended on text with no final step_finish (intermediate steps do get step_finish with tool-calls). Filed #54661 before finding this one — closing it as a duplicate.
7. oldantique

   marked run --format json: final step_finish is missing #54661 as a duplicate of this issue on Oct 12, 2026
8. sonvivu

   added a commit that references this issue on Oct 16, 2026

   fix(agent/opencode): accept clean-exit runs when OpenCode v2 omits st…

   ...

   00c5a85
9. mp-grind commented on Oct 22, 2026

   

   mp-grind

   on Oct 22, 2026

   More actions

   Just ran into this as well. Thanks for digging into it for the PR. Looking forward to seeing it get merged.

   ❤️React with ❤️1Reacted by Niels Kootstra
10. veyomr commented on Oct 27, 2026

    

    veyomr

    on Oct 27, 2026

    More actions

    Still on 2.1.4 (@opencode/cli-linux-x64) with openrouter/minimax/minimax-m3.1-flash, run --standalone --format json: final step_finish missing in 7 of 7 runs, 1 to 110 steps. Details in #59652.
11. opencodejr commented on Oct 27, 2026

    

    opencodejrBot

    on Oct 27, 2026

    More actions

    Thanks for the detailed report and for pointing to the code. I reproduced this without a real provider, using a local mock OpenAI-compatible model that first asks to run bash and then replies with text, with token usage on every response.

    • On v2.0.22, opencode run --standalone --auto --format json -m mock/m "..." | jq -r .type printed step_start tool_use step_finish step_start text in every run, with no final step_finish. Text-only replies printed just step_start text.
    • It still happens on the latest dev (fd8268ea5), run from source: the same output in every run, with and without a tool call.

    This matches your analysis. Once session.wait resolves, the client stops handling queued step events, and reconcile() only fills in text, reasoning and tool output. So the final step's finish reason, tokens and cost never reach the stream.

    The script below reproduces it, with the mock server included. #54661 and #59652 report the same problem, and the open PR #53208 targets this issue.

    Reproduction script (repro.sh)

    #!/usr/bin/env bash
    # Reproduces anomalyco/opencode#53207: `opencode run --format json` omits the final step_finish.
    # Uses an embedded mock OpenAI-compatible server (first a bash tool call, then text "tool-ok"; MOCK_TOOL=0 -> text only).
    # Needs bun, jq, git.
    # Usage: ./repro.sh [runs]   (set OPENCODE="bun ... packages/cli/src/index.ts" to run from source)
    set -u
    DIR="$(mktemp -d /tmp/oc-mock.XXXXXX)"
    cat > "$DIR/mock.ts" <<'TS'
    // Mock OpenAI-compatible chat completions server (streaming).
    // Request with no tool result yet -> bash tool call; otherwise -> text "tool-ok".
    const port = Number(process.env.MOCK_PORT ?? 4599)
    const sse = (chunks: unknown[]) =>
      chunks.map((c) => `data: ${JSON.stringify(c)}\n\n`).join("") + "data: [DONE]\n\n"
    Bun.serve({
      port,
      async fetch(req) {
        const body: any = await req.json().catch(() => ({}))
        const hasTool = (body.messages ?? []).some((m: any) => m.role === "tool")
        const wantTool = process.env.MOCK_TOOL !== "0" && !hasTool
        const base = { id: "c1", object: "chat.completion.chunk", created: 0, model: "mock" }
        const usage = { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 }
        const chunks = wantTool
          ? [
              { ...base, choices: [{ index: 0, delta: { role: "assistant", tool_calls: [{ index: 0, id: "call_1", type: "function", function: { name: "bash", arguments: "" } }] } }] },
              { ...base, choices: [{ index: 0, delta: { tool_calls: [{ index: 0, function: { arguments: JSON.stringify({ command: "echo tool-ok", description: "echo" }) } }] } }] },
              { ...base, choices: [{ index: 0, delta: {}, finish_reason: "tool_calls" }] },
              { ...base, choices: [], usage },
            ]
          : [
              { ...base, choices: [{ index: 0, delta: { role: "assistant", content: "tool-ok" } }] },
              { ...base, choices: [{ index: 0, delta: {}, finish_reason: "stop" }] },
              { ...base, choices: [], usage },
            ]
        return new Response(sse(chunks), { headers: { "content-type": "text/event-stream" } })
      },
    })
    console.log("mock on", port)
    TS
    RUNS="${1:-5}"
    OC="${OPENCODE:-opencode}"
    PROJ="$(mktemp -d /tmp/opencode/proj.XXXXXX)"; cd "$PROJ" && git init -q
    cat > opencode.json <<'JSON'
    { "providers": { "mock": { "name": "Mock", "env": ["MOCK_API_KEY"],
      "package": "@opencode/ai/providers/openai-compatible",
      "settings": { "baseURL": "http://127.0.0.1:4599/v1" },
      "models": { "m": { "name": "Mock" } } } } }
    JSON
    export MOCK_API_KEY=x
    for tool in 1 0; do
      MOCK_TOOL=$tool bun "$DIR/mock.ts" & MOCK=$!; sleep 1
      echo "== tool call: $tool"
      for i in $(seq "$RUNS"); do
        $OC run --standalone --auto --format json -m mock/m "Use bash to echo tool-ok, then reply with its output" \
          | jq -r .type | tr '\n' ' '; echo
      done
      kill $MOCK; wait $MOCK 2>/dev/null
    done

    Run it: bash repro.sh
12. opencodejr

    added

    reproduced

    on Oct 27, 2026
13. rekram1-node

    closed this as completedin #60475on Oct 30, 2026
14. wthrbtn

    added a commit that references this issue on Nov 10, 2026

    freshen(opencode-ci-harness): final step events are fixed in v2.2.3; …

    ...

    e072b3b