New Plugin

plan-agent v0.2.0

Plan authoring on demand — invoke /plan-agent:author, not on ambient intent

Manual-invoke only: /plan-agent:author <objective> — no ambient activation
Free-text objective parsing with --quick, --type, --dir, --interview flags
Smart defaults: infers type from leading verb, treats detailed objectives as --quick
EnterPlanMode entry — real plan-mode approval gate on every invocation
Filename hook unchanged — still enforces verb-target kebab-case at write time

LinkedIn post

Shipped something I've been wanting to build for a while — a plan authoring skill for Claude Code that actually requires you to ask for it.

The old `plan-mode` plugin auto-activated whenever you expressed planning intent. Helpful in theory, but in practice it kept firing when I didn't want it to. So I rethought it.

`plan-agent` v0.2.0 flips the model: planning only happens when you explicitly invoke `/plan-agent:author <objective>`. No ambient activation (`disable-model-invocation: true`), no surprise plan drafts mid-task.

What you get:
→ Free-text objective parsing with smart flag defaults
→ `--quick` skips Clarify/Align for detailed objectives
→ `--type` is inferred from your leading verb (add/fix/refactor/docs)
→ `--interview` chains into plan-interview stress-testing before approval
→ EnterPlanMode entry — real plan-mode gate, not just a workflow prompt

The filename validation hook is unchanged — it still catches random slugs and junk names at write time, exits 2 with a rename hint.

A post-ship code review also caught three real bugs in the hook: a `lstrip('./')` that treated its arg as a character set (not a prefix), a substring `in` check that triggered false positives on sibling directories, and a hex-suffix threshold mismatch. All fixed before the PR landed.

Install:
/plugin marketplace add shawn-sandy/agentics
/plugin install plan-agent@agentics-kit

#ClaudeCode #DeveloperTools #AI #Plugins