[RaycarlLei]

Description

JunieIntegration does not override build_exec_args(), so it inherits MarkdownIntegration.build_exec_args() (base.py#L1151-L1170), which passes the prompt as -p <prompt>.

In Junie CLI, -p is the short form of --project, the directory Junie runs in. A non-interactive task is given as a positional argument or with --task. From junie --help (@jetbrains/junie@3110.7.0, the npm package the JetBrains/junie README installs):

Usage: junie [<options>] [<task>]
...
Non-interactive mode:   junie "Fix the bug in the login function"
                        junie --task "Fix the bug"
...
  --task=<text>                     Task description (alternative to positional argument)
...
  -p, --project=<text>              Project directory (default: current directory)
...
  --output-format=<text>         Output format: text, json, json-stream
...
  --model=<text>                    Model to use for the primary agent

The same is in Junie's CLI reference and in the Junie skill JetBrains ships in its own repository: "-p is --project. Non-interactive is simply the positional task." (hermes-plugin/skills/junie/SKILL.md#L93).

Run inside a project created with specify init --integration junie, with HOME pointing at an empty directory (not logged in, no API key):

$ junie --version
Junie version: 26.9.7 (3110.7)
$ junie -p "/speckit-specify build a login page"
The project directory does not exist: /speckit-specify build a login page. Use --project to specify the correct path
$ echo $?
1
$ junie --task "/speckit-specify build a login page" --output-format json
{"errors":["Cannot find authorization. Please authenticate before running Junie."]}

The -p form fails before Junie looks for credentials, so this reproduces without an account. The --task form gets past option parsing and project resolution and stops only because this HOME has no authorization (Junie reports that with exit 0).

The same failure through Spec Kit on main (dc2cec4), with that junie on PATH, run from the project directory:

$ .venv/bin/python -c "from pathlib import Path; from specify_cli.integrations.junie import JunieIntegration as J; r = J().dispatch_command('speckit.plan', 'x', project_root=Path('.').resolve(), stream=False); print(r['exit_code'], r['stderr'].strip())"
1 The project directory does not exist: /speckit-plan x. Use --project to specify the correct path

On this branch the same call reaches Junie's authorization check: exit code 0, and stdout is the {"errors":["Cannot find authorization. ..."]} object above.

Reachability. Unlike the Kimi Code (#5031) and Pi (#5043) fixes, which only affected structured output, -p is passed in every mode, so the built-in workflows hit it. The command step dispatches through dispatch_command() (step/command/__init__.py#L333-L342), and the prompt step builds its argv with build_exec_args(..., output_json=False) (step/prompt/__init__.py#L248-L253); both use the inherited argv. Running the bundled workflow in the same project with a stand-in junie on PATH that only records its argv and exits 1:

$ cat "$(command -v junie)"
#!/bin/sh
# Records the argv Spec Kit passes, then exits 1 so the workflow stops at the first step.
{ printf 'argv:'; for a in "$@"; do printf ' [%s]' "$a"; done; printf '\n'; } >> "$JUNIE_SHIM_LOG"
echo "junie shim: argv recorded" >&2
exit 1
$ specify workflow run speckit -i spec="build a login page"
Running workflow: Full SDD Cycle (speckit)
Version: 1.0.1

  ▸ [specify] speckit.specify …
junie shim: argv recorded

Status: failed
Run ID: d56ce475
Error: Command exited with code 1
$ cat "$JUNIE_SHIM_LOG"
argv: [-p] [/speckit-specify build a login page]

On this branch the same run records argv: [--task] [/speckit-specify build a login page], which is the form Junie accepts above.

Not shown here, because it needs a logged-in Junie: whether a custom command passed as a non-interactive task is expanded from .junie/commands/. That also depends on #4794 (open), which reports that $ARGUMENTS in the installed commands keeps Junie from running them; this PR does not touch command content.

Fix. Override build_exec_args() in JunieIntegration to pass the prompt with --task. --task is used rather than a bare positional so the argv keeps the base shape, with the prompt right after its flag. The rest matches the base implementation: the requires_cli guard, SPECKIT_INTEGRATION_JUNIE_EXTRA_ARGS before --model, --model, and --output-format json. The full argv this branch builds, junie --task "/speckit-plan x" --skip-update-check --model sonnet --output-format json (with --skip-update-check from the extra-args variable), also gets to the authorization check above. The change stays inside the Junie integration package, its test file, and its agent page docs/integrations/junie.md#L32-L33, which named junie -p <prompt>. The new method sits at the end of the class, away from the lines the open Junie PRs (#4795, #4796, #4883, #4983) change.

Part of #2416, which lists Junie's inherited junie -p "prompt" --model X --output-format json argv as unverified. Same shape as #5031 and #5043.

Testing

• Tested locally with uv run specify --help

• Ran existing tests with uv sync && uv run pytest: run as .venv/bin/python -m pytest, results below

• Tested with a sample project (if applicable): the specify init --integration junie project above, with junie 26.9.7 (3110.7) and with the argv-recording stand-in

• tests/integrations/test_integration_junie.py gets four argv tests: text mode, --output-format json, --model, and SPECKIT_INTEGRATION_JUNIE_EXTRA_ARGS. Run against main, all four fail with '-p' != '--task'. All four pass on this branch, and the file passes as a whole (37 tests).

• .venv/bin/python -m pytest tests/integrations tests/workflows tests/test_agent_config_consistency.py -q: 3660 passed, 8 skipped.

• uvx ruff@0.15.0 check src tests: all checks passed.

• markdownlint-cli2 0.23.2 (the version the lint workflow's action pins) on docs/integrations/junie.md: 0 issues.

• Full suite, .venv/bin/python -m pytest tests -q (macOS, Python 3.14.3): 12696 passed, 434 skipped, 15 failed. All 15 are in 6 files this PR does not touch, and running those files against unmodified main on the same machine fails the same 15 test IDs. 11 are composed-template cases in test_*_python_parity.py, whose bash variant runs python3 from PATH, and that interpreter has no PyYAML on this machine; they pass with the venv's python3 first on PATH. The other 4 are in test_workflow_step_submission.py, where the fake curl gets a #!<sys.executable> shebang and the interpreter path on this machine contains a space.

AI Disclosure

• I did not use AI assistance for this contribution
• I did use AI assistance (fill in the disclosure below)

AI disclosure: Claude Code opened this PR on behalf of @RaycarlLei. Claude Code (model: Claude Opus 5.5, reasoning effort xhigh, autonomous agent mode with tool approvals bypassed) read Junie's documentation and JetBrains' Junie repository, installed @jetbrains/junie@3110.7.0 and ran it without logging in to reproduce the bug, reproduced the workflow argv with a stand-in junie, wrote the override, the four tests and the docs/integrations/junie.md update, ran the checks above, wrote this description, and pushed the branch. A read-only Claude Code subagent on the same model cross-checked the links, the reachability claims and the diff, and its corrections were applied. @RaycarlLei approved posting; he did not review the diff line by line, re-run the tests, or edit this description. The commit carries Assisted-by: Claude Code (model: Claude Opus 5.5, autonomous).

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The focused Junie-only correction preserves the existing dispatch contract, includes regression coverage, and has no unresolved findings.

No open issues 🎉

What changed in this PR

Corrects Junie’s non-interactive dispatch so prompts are treated as tasks rather than project directories, addressing part of the CLI dispatch audit in #2416.

Changes:

• Adds a Junie-specific build_exec_args() override using --task, preserving existing runtime validation and other options.
• Adds four regression tests covering prompt forwarding, JSON output, model selection, and extra arguments.
• Updates the Junie integration guide to show the corrected invocation.

Validation relies on the PR’s reported before-and-after results; tests were not rerun here.

File	Description
tests/​integrations/​test_integration_junie.py	Adds regression coverage for Junie dispatch arguments.
src/​specify_cli/​integrations/​junie/​__init__.py	Passes non-interactive prompts through --task.
docs/​integrations/​junie.md	Documents the corrected task flag.

_{Review effort: Max}

[mnriem]

Thank you!