refactor(docs): agent-readable zero-to-value onboarding (#1280)

[dolho]

Summary

Make the repo's canonical entry points machine-first so an autonomous agent landing in the repo (or connecting to an instance) can self-orient and reach a useful result with no human translating context.

Changes

• AGENTS.md — added a "Using this file" machine-contract header: declares AGENTS.md the authoritative agent entry point, states the AGENTS / CLAUDE / README boundary, explains how to traverse. Rebuilt Route by task with explicit "Done when" zero-to-value signals per persona. Links the new template catalog from the deploy section.
• CLAUDE.md — cross-link into AGENTS.md; framed CLAUDE.md as the contributor working agreement (auto-loaded by Claude Code), not the agent landing page.
• README.md — added the missing trinity agents list verify step to the CLI example.
• docs/CLI.md — leads with pip install trinity-cli (PyPI); marks -e src/cli/ as the from-source/dev variant.
• config/agent-templates/README.md (new) — machine-readable starter-template catalog: groups the 17 real templates (single-purpose + the dd-* due-diligence suite) with one-line affordances and how-to-use, and explicitly flags the 7 test/canary fixtures as "not starting points". An agent now self-selects a template instead of ls-ing 24 dirs and opening each template.yaml.

Validation (AC#5)

Ran a context-free agent through a real zero-to-value task ("deploy an agent and verify it worked") using only repo files, no human context:

• Self-oriented in 2 hops (README → AGENTS.md) to a correct deploy + verify answer.
• Verdict: "the routing itself is excellent and is the model other repos should copy."
• Its friction log produced four doc fixes (undefined $TOKEN, conflicting CLI install, missing README verify, unstated deploy prerequisite) — folded back in.

Acceptance criteria

• Audit agent-facing entry points for readability gaps (via the validation run)
• Define zero-to-value per persona (explicit "Done when" signals)
• Restructure / annotate entry points so an agent self-orients (task router + boundary + template catalog)
• Establish AGENTS.md as the authoritative entry point that cross-links the rest
• Validate with an agent performing a representative task using only repo files

Notes

• Human docs left intact (agent-first defaults, not removal of human readability).
• Repo-root AGENTS.md is hand-authored; the CLAUDE.md → AGENTS.md mirror (Codex harness MVP — pluggable agentic execution engine alongside Claude Code #1187) is per-agent-container (startup.sh), so these repo-root edits are conflict-free.
• The .claude/ skills affordance pass is deliberately not here: skills live in the private trinity-dev submodule and can't land in a public-repo PR — tracked separately.

Related to #1280

🤖 Generated with Claude Code

[github-actions]

⚠️ Nightly unit-suite check skipped — merge conflict against dev.

Resolve by running git merge dev locally and pushing the result. The next nightly run will re-test once the conflict is gone.

[vybe]

Validated via /validate-pr: docs-only, base=dev, conventional commits, security-clean (no secrets/emails/IPs), enterprise-docs guard passed. Approving.