Add BASECAMP_NONINTERACTIVE escape hatch for prompts by robzolkos · Pull Request #520 · basecamp/basecamp-cli

robzolkos · 2026-07-03T02:15:03Z

Problem

An agent that runs basecamp todos list --in <id> --md can be blocked by an interactive selection prompt (e.g. picking a todoset when a project has more than one), wedging the session. Reported in #395.

The root cause is that --md is deliberately not a machine-output mode — only --agent/--json/--quiet/--ids-only/--count suppress prompts (IsInteractive() in both internal/appctx/context.go and internal/tui/resolve/resolve.go). When the CLI runs under an allocated PTY, stdout looks like a terminal, so ambiguous resolutions fall through to a blocking picker even though a human isn't there to answer it.

Relying on the model to remember --agent per invocation is exactly what the issue asks us to avoid.

Change

Add BASECAMP_NONINTERACTIVE as an explicit escape hatch. When set to a truthy value (1/true), it forces IsInteractive() to false in both gates, so ambiguous resolutions become actionable errors (Multiple todosets found — specify one with --todoset <id>) instead of prompts.

Command-level prompt guards also consult the escape hatch, so direct confirmations such as basecamp chat delete ... without --force cannot wedge an agent session under a PTY.

Key property: it only disables prompts — it does not change the output format. That is the distinction from --agent (which also forces quiet JSON): you keep --md Markdown output and just lose the wedge.

Setting it is the environment/harness's responsibility (shell profile, MCP server env, wrapper), not the model's — so it doesn't depend on the LLM appending a flag every call. Mirrors the pattern the issue references from other agent CLIs.

Details

config.NonInteractiveEnv() reuses the existing strict parseEnvBool (accepts 1/true, ignores unrecognized values).
Both IsInteractive() chokepoints consult it first; resolver prompts and app.IsInteractive() command sites flow through those gates.
Command-level prompt guards also consult it, covering direct confirmations that do not use the resolver.
Explicit setup flows (basecamp setup and interactive skill/setup wizards) remain intentionally interactive; use direct flags/subcommands for non-interactive setup.
skills/basecamp/SKILL.md now documents that --md does not suppress prompts and points to the env var / disambiguating flags.

Notes for reviewers

The env var's effect on IsInteractive() is inherently TTY-dependent, so the unit tests assert the parser (config) and the invariant that the escape hatch does not flip IsMachineOutput() — matching the limitation of the existing IsInteractive tests.
This does not touch the pre-existing --agent behavior.

Refs #395

Summary by cubic

Add a BASECAMP_NONINTERACTIVE env escape hatch to disable prompts without changing output format, so --md runs under a PTY don’t get stuck. Also treats --jq as machine-output for the profile picker and clarifies character-device detection in docs.

New Features
- Set BASECAMP_NONINTERACTIVE to 1 or true to force non-interactive mode across the app, resolver, profile picker, and command prompts; ambiguous targets return clear errors (e.g., use --todoset <id>).
- Output format is unchanged; you can keep --md while suppressing prompts. --jq implies JSON and now suppresses the profile picker.
- Docs updated to note stdout “character device (e.g., a terminal)” detection and fix disambiguation examples in skills/basecamp/SKILL.md; tests hardened to cover the env short-circuit, char-device stdout, and command prompt guards.

^{Written for commit 58180e5. Summary will update on new commits.}

Only the machine-output flags (--agent/--json/--quiet/--ids-only/--count) suppress interactive selection prompts. --md does not, so an agent that requests Markdown output and runs under an allocated PTY (where stdout looks like a terminal) can be wedged by a blocking picker when a target is ambiguous — e.g. a project with multiple todosets and no --todoset. Add BASECAMP_NONINTERACTIVE as an explicit escape hatch. When set to a truthy value it forces IsInteractive() to false in both gates (appctx and resolve), turning ambiguous resolutions into actionable errors instead of prompts — without changing the output format the way --agent does. Setting it is the environment/harness's responsibility, not the model's, so it does not rely on the LLM remembering a flag per invocation. Refs #395

Copilot

Pull request overview

Adds a BASECAMP_NONINTERACTIVE environment-variable escape hatch so agents/harnesses can reliably disable interactive selection prompts (even under a PTY) without switching output formats (e.g., keep --md Markdown output).

Changes:

Introduce config.NonInteractiveEnv() to strictly parse BASECAMP_NONINTERACTIVE (1/true) as a non-interactive override.
Apply the override to both interactive “chokepoints”: internal/appctx.App.IsInteractive() and internal/tui/resolve.Resolver.IsInteractive().
Document prompt-suppression behavior and the new env var in skills/basecamp/SKILL.md, and add unit tests for the env parsing / invariants.

Tip

If you aren't ready for review, convert to a draft PR.
Click "Convert to draft" or run gh pr ready --undo.
Click "Ready for review" or run gh pr ready to reengage.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
skills/basecamp/SKILL.md	Documents that `--md` doesn’t suppress prompts and adds `BASECAMP_NONINTERACTIVE` guidance.
internal/tui/resolve/resolve.go	Disables resolver prompts early when `BASECAMP_NONINTERACTIVE` is truthy.
internal/config/config.go	Adds `NonInteractiveEnv()` using existing strict `parseEnvBool`.
internal/config/config_test.go	Tests env parsing behavior for `BASECAMP_NONINTERACTIVE`.
internal/appctx/context.go	Disables app-level interactivity early when `BASECAMP_NONINTERACTIVE` is truthy.
internal/appctx/context_test.go	Adds an app-level test asserting the escape hatch doesn’t change machine-output mode.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

cubic-dev-ai

3 issues found across 6 files

_{Reply with feedback, questions, or to request a fix.

Re-trigger cubic}

- context_test: swap stdout to /dev/null (a char device) so the test actually exercises the BASECAMP_NONINTERACTIVE short-circuit instead of passing because go test's stdout is a pipe. Baseline require.True asserts the interactive path, then the env var flips it to false. - config_test: use t.Setenv("") for the unset case instead of os.Unsetenv, so testing.T restores the prior value (no env leak into other tests). - SKILL.md: include BASECAMP_NONINTERACTIVE in the prompt-suppression list to remove the "Only <flags> suppress prompts" contradiction.

robzolkos · 2026-07-03T02:24:42Z

Addressed all three review findings in 986e96c:

config_test env leak — the unset case now uses t.Setenv("") (which NonInteractiveEnv treats as unset) instead of os.Unsetenv, so testing.T restores the prior value.
context_test not exercising the short-circuit — the test now swaps os.Stdout to /dev/null (a char device), asserts the interactive baseline with require.True, then sets BASECAMP_NONINTERACTIVE=1 and asserts IsInteractive() flips to false. It now fails if the short-circuit is removed.
SKILL.md contradiction — the sentence now lists the flags and BASECAMP_NONINTERACTIVE=1 together as prompt suppressors.

Full test suite green against the pinned SDK (GOWORK=off go test ./...).

Copilot

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

cubic-dev-ai

2 issues found across 2 files (changes from recent commits).

_{Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic}

- root.go: isInteractiveTTY now treats --jq (JQFilter) as machine output, matching App.IsInteractive. Without it, --jq + multiple profiles could still hit the interactive profile picker despite --jq implying --json. - root_test.go, context_test.go: use os.DevNull instead of a hard-coded "/dev/null" so the tests stay meaningful on non-Unix platforms.

Copilot

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

- resolve.go, root.go: doc/inline comments said "terminal", but the guard checks os.ModeCharDevice, which also matches non-terminal char devices like /dev/null (the tests rely on this). Reword to say "character device (e.g. a terminal)". - SKILL.md: the disambiguation list implied --in <project> resolves the "multiple todosets" ambiguity. Reword so each flag maps to the ambiguity it actually resolves.

Copilot AI review requested due to automatic review settings July 3, 2026 02:15

github-actions Bot added tui Terminal UI tests Tests (unit and e2e) skills Agent skills enhancement New feature or request labels Jul 3, 2026

Copilot started reviewing on behalf of robzolkos July 3, 2026 02:15 View session

Copilot AI reviewed Jul 3, 2026

View reviewed changes

Comment thread internal/config/config_test.go Outdated

Comment thread internal/appctx/context_test.go

Comment thread skills/basecamp/SKILL.md Outdated

cubic-dev-ai Bot reviewed Jul 3, 2026

View reviewed changes

Comment thread internal/appctx/context_test.go

Comment thread skills/basecamp/SKILL.md Outdated

Comment thread internal/config/config_test.go Outdated

Honor noninteractive env for profile picker

6dc8510

Copilot AI review requested due to automatic review settings July 3, 2026 02:44

Copilot started reviewing on behalf of robzolkos July 3, 2026 02:44 View session

Copilot AI reviewed Jul 3, 2026

View reviewed changes

Comment thread internal/cli/root.go

Comment thread internal/cli/root_test.go Outdated

Comment thread internal/appctx/context_test.go Outdated

cubic-dev-ai Bot reviewed Jul 3, 2026

View reviewed changes

Comment thread internal/cli/root.go Outdated

Comment thread internal/cli/root_test.go Outdated

robzolkos added 2 commits July 2, 2026 23:09

Honor noninteractive env for command prompts

ae27765

Copilot AI review requested due to automatic review settings July 3, 2026 03:38

github-actions Bot added the commands CLI command implementations label Jul 3, 2026

Copilot started reviewing on behalf of robzolkos July 3, 2026 03:38 View session

Copilot AI reviewed Jul 3, 2026

View reviewed changes

Comment thread internal/tui/resolve/resolve.go Outdated

Comment thread internal/cli/root.go Outdated

Comment thread skills/basecamp/SKILL.md Outdated

robzolkos merged commit f9041ac into main Jul 3, 2026
26 checks passed

robzolkos deleted the noninteractive-env-escape-hatch branch July 3, 2026 03:50

robzolkos mentioned this pull request Jul 3, 2026

Agent not passing --agent when requesting --md output mode #395

Closed

Uh oh!

Conversation

robzolkos commented Jul 3, 2026 • edited by cubic-dev-ai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Change

Details

Notes for reviewers

Summary by cubic

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

robzolkos commented Jul 3, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

robzolkos commented Jul 3, 2026 •

edited by cubic-dev-ai Bot

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading