fix: route non-builtin slash commands after TUI dispatch

[deseltrus]

TL;DR

What: Restore non-builtin slash-command routing in the interactive TUI so extension commands, prompt templates, and enabled /skill:* inputs are not rejected as unknown.
Why: The unknown-command fallback added in the TUI slash dispatcher consumed any unmatched /... before session-level slash resolution could run, and /export also matched unrelated /export* prefixes.
How: Move unknown-command rejection to the interactive entry points that can distinguish builtins from session-resolved slash commands, gate /skill:* recognition on the skill-command setting, and tighten /export matching to exact command tokens.

What

This change updates the interactive slash-command routing path in pi-coding-agent.

File	Change
packages/pi-coding-agent/src/modes/interactive/slash-command-handlers.ts	Keep the dispatcher focused on true builtins and require exact /export tokens instead of matching every /export* prefix.
packages/pi-coding-agent/src/modes/interactive/controllers/input-controller.ts	Reject only truly unknown slash commands after builtin dispatch, while allowing known session-resolved slash commands to continue to session.prompt().
packages/pi-coding-agent/src/modes/interactive/interactive-mode.ts	Reuse the same known-slash classification for follow-up and compaction queue entry points, and only recognize /skill:* when skill commands are enabled.
packages/pi-coding-agent/src/modes/interactive/controllers/input-controller.test.ts	Add regression coverage for extension commands, prompt templates, enabled and disabled skill commands, /export prefix collisions, and unknown slash commands.

Why

dispatchSlashCommand() started treating every unrecognized /... input as handled inside the TUI before session.prompt() could run its normal slash resolution. That meant valid non-builtin slash inputs never reached extension command execution, prompt-template expansion, or /skill:* expansion.

It also left a narrower routing bug in place because /export matched any /export* prefix, which could still steal unrelated slash names from later resolution.

Fixes #3421
See also #3422

How

The fix keeps builtin handling in the TUI, but moves unknown-command rejection to the interactive entry points that know when a slash input should fall through to session-level routing:

• builtin commands still execute directly in dispatchSlashCommand();
• extension commands, prompt templates, and enabled /skill:* commands continue to session.prompt();
• truly unknown slash inputs show the error before they are queued or sent;
• /export now matches only /export and /export ..., not arbitrary /export* prefixes.

This stays within the existing command model rather than adding a new routing layer.

Test Evidence

• node --import ./src/resources/extensions/gsd/tests/resolve-ts.mjs --experimental-strip-types --test packages/pi-coding-agent/src/modes/interactive/controllers/input-controller.test.ts
• npm run build

---

AI-assisted contribution — reviewed and tested by contributor.

• fix

[github-actions]

🟠 PR Risk Report — HIGH

Files changed	4
Systems affected	3
Overall risk	🟠 HIGH

Affected Systems

Risk	System
🟠 high	Modes
🟡 medium	TUI Components
🟡 medium	Commands

File Breakdown

Risk	File	Systems
🟠	packages/pi-coding-agent/src/modes/interactive/controllers/input-controller.ts	Modes, TUI Components
🟠	packages/pi-coding-agent/src/modes/interactive/interactive-mode.ts	Modes, TUI Components
🟠	packages/pi-coding-agent/src/modes/interactive/slash-command-handlers.ts	Modes, Commands
⚪	packages/pi-coding-agent/src/modes/interactive/controllers/input-controller.test.ts	(unclassified)

⚠️ High risk — please run full integration tests and verify tool/extension contracts.

[justinwyer]

@jeremymcs I see the publish pipeline is stuck waiting on a pipeline from yesterday

[jeremymcs]

Sorry guys. I stepped away this morning for a golf tourney. Didn't see stuck pipeline.

[justinwyer]

@jeremymcs thanks so much hope you beat your handicap!

[TerminalSausage]

i don't think this release was pushed out for npm installation. Latest is still pushing the broken version. Thanks for your work on this project.

[jeremymcs]

Sorry for the long delay in getting NPM up guys. We've got a gap in permissions and @glittercowboy is waiting on some hardware replacments to get back online. Hopefully he'll be able to jump on tonight and get 2.60 pushed out to NPM.