foundation: manifesto, origin, governance templates, and Phase 0 planning docs

[EmersonBraun]

Summary

First concrete step of Phase 0 — Foundation Hardening (PRD #211). Establishes narrative, governance baseline, and planning documents that unblock all subsequent work.

What's included

Narrative foundations (root)

• MANIFESTO.md — 10 non-negotiable principles (core <10KB, plug-and-play, interop, zero lock-in, agent-first, docs-as-product, TS rigor) plus explicit "what we'll never/always do" lists
• ORIGIN.md — first-person origin story explaining why AgentsKit exists and the bet it's making

Governance (.github/)

• CODEOWNERS — per-package ownership with maintainer fallback (extensible as team grows)
• ISSUE_TEMPLATE/rfc.yml — structured RFC template with Manifesto-alignment checklist
• ISSUE_TEMPLATE/docs.yml — dedicated docs issue template
• ISSUE_TEMPLATE/config.yml — disables blank issues, routes questions to Discussions
• PULL_REQUEST_TEMPLATE.md — expanded with type, RFC reference, manifesto checklist, DoD aligned with Phase 0 gates

Planning documents (docs/)

• ROADMAP-PRD.md — Master PRD with 97 user stories across 4 phases (ref [Master PRD] AgentsKit Roadmap — 4 phases, 97 user stories #113)
• PHASE-0-FOUNDATION-PRD.md — 52 stories of foundation hardening (ref [Master PRD] Phase 0 — Foundation Hardening (antes da Fase 1) #211)
• PHASE-0-EXECUTION-PLAN.md — tactical P0/P1/P2 plan with day-by-day sprint 1
• MASTER-EXECUTION-PLAN.md — 149-issue sequencing across 27 sprints over ~12 months (ref [Master] Execution Plan — 149 issues em 12 meses #264)

Why this first

These documents are zero-code dependencies for everything else in Phase 0:

• CODEOWNERS enables meaningful review flow on the 149 open issues
• RFC template unblocks the 6 ADRs for core contracts ([P0.3] ADRs for 6 core contracts (Adapter/Tool/Memory/Retriever/Skill/Runtime) #214)
• Manifesto gives a reference point every future PR can be measured against
• Planning docs let contributors pick work aligned with the sequencing

Test plan

• MANIFESTO.md and ORIGIN.md render correctly on GitHub
• CODEOWNERS validated by GitHub (no syntax errors)
• New issue templates appear when opening a new issue
• PR template used by this very PR
• No code changes — pure documentation
• Reviewer: confirm principles in MANIFESTO reflect project intent
• Reviewer: confirm ORIGIN tone is authentic

Follow-ups (separate PRs)

• ARCHITECTURE.md and first ADR 0001-adapter-contract.md ([P0.3] ADRs for 6 core contracts (Adapter/Tool/Memory/Retriever/Skill/Runtime) #214)
• agentskit.io DNS setup ([P0.24] Configure agentskit.io (DNS + SSL + redirect) #235)
• size-limit bundle gate pilot in @agentskit/core ([P0.6] Bundle size budget enforced in CI (size-limit) #217)
• Fumadocs migration spike ([P0.27] Decision Tree (5 questions → packages) #238)

Refs #113 #211 #264

[EmersonBraun]

Test fix update

CI test failure was a pre-existing regression on main (not introduced by this PR):

• ink was bumped ^6.8.0 → ^7.0.0 on main, but ink-testing-library@4 (latest) doesn't route stdin through ink v7's new raw-mode pipeline, so useInput callbacks never fire under test.
• Verified by running the same tests on main — they fail identically.

What I did

• Skipped the 5 affected tests with it.skip + TODO(#266) comments linking to the tracking issue.
• Component code is untouched — only test harness is bypassed.

Verification

pnpm test (all 22 tasks) now passes locally. ink package reports: 16 passed | 5 skipped.

Tracking

• Issue [Phase 0] ink-testing-library incompatible with ink@^7 — useInput stdin events #266 tracks the proper fix (contribute to ink-testing-library, write custom harness, or downgrade ink).

This is medium-priority: the terminal components work correctly in real use; only automated coverage of keyboard-input scenarios is temporarily reduced.

[EmersonBraun]

Tests fixed (no regressions)

Updated approach: instead of skipping, rewrote the 5 failing tests to work around the ink@7 / ink-testing-library@4 incompat properly.

Root cause

Ink v7 moved useInput off direct stdin listening onto an internal EventEmitter that routes through a readable-event input parser. ink-testing-library@4's mocked Stdin emits 'readable' but the pipeline no longer picks up the handler in a way that triggers useInput. Verified by writing a direct repro outside vitest — real stdin.write(...) on the library mock never fires useInput under ink 7.

Fix

Mock ink's useInput in test files only: capture the handler, then drive it directly with the same (input, key) shape ink 7 passes. This tests the actual component logic (setInput/send/guards) without depending on the broken harness.

Results

• @agentskit/ink: 23 passed, 0 skipped (was 16 passed, 5 skipped)
• All 22 workspace test tasks green
• Two new coverage cases added (ctrl and meta key combos ignored)

Issue #266 closed by this commit.

Unrelated: docs build

@agentskit/docs (Docusaurus 3.9 + Node 25) also fails on main with mdxCrossCompilerCache error — pre-existing, not touched by this PR. Will be resolved naturally by the Fumadocs migration (#238).