v0.1.269 — agent-deployments first release

First tagged release of agent-deployments — composable agent blueprints for production AI deployments, consumed by the agent-scaffold CLI.

What's in this release

A complete, machine-readable, internally-consistent spec catalog for building production-shaped AI agents:

• 10 recipe blueprints — every recipe ships a canonical YAML frontmatter (status, languages, required_files, recipe_dependencies, external_services, capabilities, bootstrap_config, topology, roles[]), a ## Composes opener with a ### Load list inside, and the 13-section body shape documented in docs/recipes/SCHEMA.md. 4 are validated (with reference implementations); 6 are design-spec.
• 16 capability files across 8 kinds (vector_db, cache, relational, queue, obs, eval, frontend, host) — every one conforms to the schema in docs/capabilities/README.md, with pinned image tags, declared bootstrap steps, and resolvable emit_files: source paths.
• 22 cross-cutting docs covering auth, logging, observability, rate limiting, idempotency, DLQ ops, multi-tenancy, cost tracking, model routing, schema evolution, and more — referenced from recipes via prose, threaded into the LLM context tier by the scaffold.
• Template trees for host.vercel, host.railway, host.fly, frontend.nextjs-chat, frontend.streamlit, obs.grafana-stack, and eval.promptfoo — runnable scaffolds that the agent-scaffold emitter copies into generated projects per the emit_files: contract.
• Canonical stack docs for 19 infrastructure picks (Postgres, Redis, Qdrant, Kafka, Langfuse, FastAPI, Hono, Vercel AI SDK, Pydantic AI, LangGraph, Mastra, CrewAI, etc.) — 7 of them carry bi-directional > Capability: cross-links back into the catalog.

Highlights of the immediate run-up to this release (PRs #35 → #40)

#	Title	Scope
#35	feat(recipes): publish canonical recipe-frontmatter schema	New docs/recipes/SCHEMA.md; align indexes; cross-link 3 capability docs
#36	feat(recipes): frontmatter + canonical structure on validated recipes	Retrofit 4 anchor recipes (docs-rag-qa, customer-support-triage, research-assistant, restaurant-rebooking); add missing Eval Datasets
#37	feat(recipes): frontmatter + canonical structure on design-spec recipes	Retrofit 6 design-spec recipes (memory-assistant, content-pipeline, parallel-enricher, ops-crew, code-review-agent, hierarchical-agent)
#38	feat(capabilities): host capability templates + template-tree convention doc	Vercel/Railway/Fly stubs + new TEMPLATES.md
#39	feat(capabilities): eval.deepeval and eval.ragas capability stubs	Close the eval-catalog asymmetry; convert promptfoo swap targets to live links
#40	chore(stack): remove legacy stack.md duplicate; complete capability cross-links	Delete docs/stack.md; finish bi-directional cross-links

Verification at release time

• 634 internal markdown links checked, 0 broken.
• All 10 recipes declare both external_services: (v0.2.x reach) and capabilities: (v0.3+ reach).
• All capability ids declared across all 10 recipes resolve to existing files under docs/capabilities/<kind>/<name>.md.
• All emit_files: source: paths declared across all 16 capabilities resolve to existing files.
• Zero build artifacts in the tree; .gitignore covers __pycache__/, *.pyc, *.tsbuildinfo, node_modules/, .next/, dist/, build/.
• Zero leaked local paths or TODO/TBD markers.

Compatibility with agent-scaffold

• v0.2.x (live, v0.2.255 verified) — parses status, languages, required_files, recipe_dependencies, external_services from recipe frontmatter. Every recipe declares all five.
• v0.3+ (pending, "Track C") — additionally parses capabilities, bootstrap_config, topology, roles[]. The additive fields are silently ignored on v0.2 per docs/capabilities/README.md, so recipes work against both versions today.

Known nits (deferred to follow-up)

• Stack-doc Capability headers use two formats: 6 docs use the verbose > **Capability:** [...] (provisioning contract for agent-scaffold up). Alternatives: ..., and 1 (stack/eval-deepeval-ragas-promptfoo.md) uses the simpler > Capabilities: A · B · C. Both render and link correctly; normalization is a one-line follow-up if desired.
• docs/quickstart.md partly overlaps with the root README.md Quick Start section. The two drift independently. Stylistic, no consumer impact.
• A CI schema-conformance linter for recipe frontmatter against docs/recipes/SCHEMA.md is the natural follow-up to the schema PR but not included in this release. Drift will reintroduce silently without it.

Cross-repo coordination flags (for the scaffold team)

These are markdown-side changes that need v0.3-side confirmation before recipes lean harder on them:

1. bootstrap_config: top-level key — confirm v0.3 loader reads it under that name.
2. roles[].model_fallbacks and roles[].cost_budget_usd_per_day — runtime concerns referenced by cross-cutting/model-routing.md and cross-cutting/cost-tracking.md.
3. role_kind enum (supervisor | worker | router | notifier) — confirm validation or opaque treatment.
4. eval.deepeval / eval.ragas ids — verify resolver is by-id lookup, not by-enum.
5. required_files semantics — confirm whether v0.3 enforces or just suggests.

Counting convention

v0.1.269 follows the user's chosen convention: 0.1.<git rev-list --count main> for this first release. Subsequent releases can adopt semver patch numbers or stay on the commit-count convention as preferred.