feat(plan): tighten compact task index contract (T2-T7)

[vipentti]

Summary

Implements compact-task-index-contract T2-T7 per plans/compact-task-index-contract/plan.md. Tightens planning guidance from permissive 60/100-word wording to 25/50-word targets and adds explicit pre-presentation compression pass. Names compression pass before proposal presentation in planlet-plan skill, deferring operational detail to guidance. Regenerates tracked installed skill copies and both .planlet-manifest.json files from freshly built dist/planlet.mjs. Adds structural contract assertions for compact template properties and new guidance controls. Adds [Unreleased] changelog entry for published-skill behavior change. Completes planlet via normal lifecycle flow. Revises task-boundary guidance so independently meaningful delivered outcomes are the sole split criterion, folded into existing split-rather-than-compress rule as one boundary rule.

Excluded scope preserved: no src/ parser/validator changes or task syntax, no assets/tasks-template.md rewrite, no planlet_design.md edit, no new standard named metadata field, no planlet-implement or planlet-complete skill changes, no existing-plan migration.

CLI behavior

• Commands/options: none changed; uses existing node dist/planlet.mjs update and node dist/planlet.mjs --root . tools outputs plus node dist/planlet.mjs complete
• Output or errors: none
• Compatibility: guidance targets are planner judgment with no parser or validator enforcement; existing planlets remain valid

Implementation notes

• Planning-guidance-only scope for task-boundary rules; SKILL.md unchanged for this rule per review request.
• Revised from previous Prefer several focused tasks... split when parts can be implemented, reviewed, or verified separately to: independently meaningful delivered outcomes as sole criterion, a task may contain several testable/reviewable components when they serve one coherent outcome (requirements stay in plan.md), split only when every resulting checkbox is its own coherent outcome (splitting rather than compressing), separate implementability/verifiability is only evidence never sufficient, word-count pressure must not change semantic boundaries, anti-mechanical guard preserved — folded into single task-boundary rule rather than overlapping definitions.
• Guidance ownership and existing optional Verify: behavior preserved per plan exclusions.
• Installed copies regenerated in correct order: npm run build first (fresh checkout has no built CLI), then node dist/planlet.mjs update, then node dist/planlet.mjs --root . tools all 4 installed and both manifests in same changeset (this follow-up and prior follow-up).
• Previous follow-ups: refactored tests/skills/skill-contract.test.ts to paragraph/step-scoped relationship proofs; this follow-up adjusts structural test to prove every resulting checkbox names its own coherent delivered outcome relationship without brittle exact-prose matching (single boundary paragraph: sole criterion, several requirements→one outcome+keep in plan.md, every checkbox→coherent outcome+splitting rather than compressing, only evidence/never sufficient, word-count pressure+semantic boundaries, anti-mechanical).

Verification

• npm run format:check
• npm run lint
• npm run knip
• npm run type-check
• npm run build
• npm test (250 pass, 0 fail)
• git diff --check
• node dist/planlet.mjs --root . validate compact-task-index-contract → valid, completed 6/6
• node dist/planlet.mjs --root . status compact-task-index-contract → completed
• node dist/planlet.mjs --root . tools → all 4 destinations installed
• node dist/planlet.mjs complete compact-task-index-contract → 2026-08-17-compact-task-index-contract, completedAt: 2026-08-17T08:57:01.957Z, mode normal, no override (retained)
• git status --short no unexpected paths; plan remains completed
• no src/ changes verified

Contributor checklist

• Tests cover new or changed behavior, or no test change is needed.
• User-facing documentation and CLI help are updated, or no update is needed.
• CHANGELOG.md: [Unreleased] entry added for user-visible changes in the same change set, or explicit skip reason stated for internal-only work.
• Every planlet task this PR completes is checked with planlet task check <slug> <task-id>, and a plan whose last task this PR completes is closed with planlet complete <slug>; or the PR completes no planlet task.
• Canonical skill changes have been regenerated with node dist/planlet.mjs update --tools all, or no skill changed.
• The change preserves supported Node.js and operating-system compatibility.

Related work

• Implements plans/compact-task-index-contract/ (PR docs(plan): compact tasks.md execution index contract #93 follow-up), now plans/completed/2026-08-17-compact-task-index-contract/
• Tasks: T2 T3 T4 T5 T6 T7 checked; plan completed normally (no incomplete override)