Skip to content

Releases: aihxp/deploy-ready

v1.0.11

06 May 12:56

Choose a tag to compare

Documentation-only patch. Adds first-class support for pi and OpenClaw via the Agent Skills standard. The suite now positions explicitly as agentskills.io-compatible: any harness that parses SKILL.md frontmatter natively runs every ready-suite skill first-class, with no per-tool integration. pi and OpenClaw both load skills from the neutral ~/.agents/skills/ path defined by the standard, so future AgentSkills-compatible harnesses inherit support for free.

Changed

  • Frontmatter: compatible_with now lists pi, openclaw, and any-agentskills-compatible-harness (the latter replaces the older any-agent-with-skill-loading value for tighter standards-level signaling).
  • SUITE.md: install-locations table adds rows for pi, OpenClaw, and the neutral Agent Skills path; new "Standards" section names the standard and the verified harnesses (Claude Code, Codex, Cursor, Windsurf, pi, OpenClaw).
  • Hub install.sh / uninstall.sh (in aihxp/ready-suite): detect pi (~/.pi/) and OpenClaw (~/.openclaw/), write to the neutral ~/.agents/skills/ path. No regressions on the existing Claude Code / Codex / Cursor flow.

Why a patch, not a minor

The skill's behavior, references, and workflow are unchanged. Only the frontmatter compatible_with list, SUITE.md, and the README install section move; the version + updated frontmatter fields tick. Per the suite versioning discipline, patch-level is the honest bucket for documentary-only standards-compliance signaling.

v1.0.10

06 May 16:10

Choose a tag to compare

Documentation-only patch. Suite-wide SUITE.md refresh introducing the new orchestration tier and the eleventh sibling, kickoff-ready v1.0.0. kickoff-ready sequences the ten core-suite specialists for greenfield projects from raw user intent: it triggers from a fresh idea ("kickoff," "I have an idea help me ship it") and invokes prd-ready -> architecture-ready -> roadmap-ready -> stack-ready -> repo-ready -> production-ready -> deploy-ready -> observe-ready -> (launch-ready || harden-ready), verifying each artifact on disk before advancing. It produces only .kickoff-ready/PROGRESS.md; it never produces specialist content. No behavioral changes to this skill.

Changed

  • SUITE.md: new "orchestration" tier introduced as the first column of the four-tier diagram; new kickoff-ready row in the per-skill table; updated dependency-flow text; new composition principle 8 ("Orchestration is one-way: kickoff-ready knows about specialists; specialists do not know about kickoff-ready"); known-good versions table now lists eleven skills.

Why a patch, not a minor

The skill's behavior, frontmatter contract, and reference library are unchanged. Only the cross-sibling SUITE.md is touched and the version + updated frontmatter fields tick. Per the suite versioning discipline, patch-level is the honest bucket for documentary-only sibling-ship tracking.

v1.0.9

29 Apr 20:06

Choose a tag to compare

Documentation-only patch. Removes the Mirror Box dogfood track from the ready-suite. The aihxp/mirror-box repo has been archived; the canonical-dogfood section in SUITE.md is gone; per-skill dogfood/ folders are deleted.

Changed

  • SUITE.md no longer carries the "Canonical dogfood target: Mirror Box" section. Byte-identical sync across all ten siblings.
  • dogfood/ folder removed from this repo.

Why

The Mirror Box reference implementation required real infrastructure (Fastify + OTel + Fly.io + Honeycomb account) to fully exercise. The user wanted a skill suite, not a project that demands a particular hosted stack. Removing Mirror Box restores that posture: every skill stands on its own SKILL.md plus references; downstream consumers compose via the artifact contracts the skills describe, without depending on a shared hosted exemplar.

The interop standard is unchanged. Skills still produce .{skill}-ready/*.md artifacts; downstream siblings still read them. The contract holds without a canonical demo.


v1.0.8

24 Apr 15:02

Choose a tag to compare

Documentation-only patch. Suite-wide SUITE.md refresh introducing the canonical dogfood target: aihxp/mirror-box. Adds a new "Canonical dogfood target" section to SUITE.md with links to the ten per-skill dogfood artifacts. Adds composition principle #7 codifying the byte-identical-SUITE.md invariant across siblings. No behavioral changes to the skill.

Changed

  • SUITE.md: new "Canonical dogfood target: Mirror Box" section with artifact links.
  • SUITE.md: composition principles now include #7 (byte-identical SUITE.md across siblings).
  • SUITE.md: version table bumped; all ten skills reflect the coordinated sync.

Why a patch, not a minor

Same rationale as prior x.y.z patches: the skill's behavior, frontmatter contract, and reference library are unchanged. Only the cross-sibling SUITE.md is touched. Per the suite versioning discipline, patch-level is the honest bucket for documentary-only sibling-ship tracking.

v1.0.7

23 Apr 22:37

Choose a tag to compare

Documentation-only patch. Reflects the arrival of harden-ready v1.0.0 (https://github.com/aihxp/harden-ready) as a live sibling in the ready-suite. harden-ready is the tenth and final core-suite skill; its v1.0.0 release completes the shipping tier alongside deploy-ready, observe-ready, and launch-ready, and completes the ready-suite across planning (four), building (two), and shipping (four) tiers. harden-ready owns post-deploy adversarial review, OWASP Top 10 walkthroughs (Web / API / LLM), compliance control-to-code mapping (SOC 2 / HIPAA / PCI-DSS / GDPR), pen-test preparation and retest discipline, responsible-disclosure program design beyond SECURITY.md, and class-not-instance post-incident hardening. No behavioral changes to this skill.

Changed

  • SUITE.md updated to list harden-ready at 1.0.0 alongside the coordinated one-patch bump across every live sibling. Copy remains byte-identical across every live sibling.
  • SKILL.md frontmatter version bumped to 1.0.7. No content change beyond the version tag.

Why a patch, not a minor

The skill's behavior, frontmatter contract, and reference library are unchanged. Only the cross-sibling SUITE.md is touched to track a new sibling's release. Per the suite versioning discipline, patch-level is the honest bucket for documentary-only sibling-ship tracking.

v1.0.3 Sibling sync: launch-ready 1.0.0 live

23 Apr 07:38

Choose a tag to compare

Documentation-only patch. SUITE.md updated to reflect launch-ready v1.0.0. Shipping tier is now complete.

v1.0.2 Sibling sync: repo-ready 1.6.0 live

23 Apr 07:25

Choose a tag to compare

v1.0.2 (2026-04-23)

Documentation-only patch. Reflects the arrival of repo-ready v1.6.0 as a live sibling in the ready-suite with its suite-membership retrofit (frontmatter interop fields, SUITE.md, Unicode cleanup). No behavioral changes to the skill.

Changed

  • SUITE.md known-good versions table updated: repo-ready now shows version 1.6.0 and its repo URL instead of "See its CHANGELOG."
  • SKILL.md frontmatter version bumped to 1.0.2. No content change beyond the version tag.

v1.0.0 Initial release

23 Apr 07:29

Choose a tag to compare

v1.0.0 (2026-04-22)

First stable release of deploy-ready, the shipping-tier skill that owns the pre-prod-to-prod handoff in the ready-suite. Ships with the full SKILL.md contract, ten reference files, a ~5000-word research report backing every guardrail, and full interop-standard frontmatter. Dogfooded against a realistic solo-dev Fly.io deploy before cut; the rough edges surfaced are reflected in the refinements below.

The skill's three named failure modes

deploy-ready introduces three terms the ecosystem did not already have a clean name for. Each maps to a specific class of real-world incident (citations in references/RESEARCH-2026-04.md).

  • Paper canary. A canary without a named success metric, a numeric threshold, a time or request window, and an automated rollback trigger. Appears green because nothing is looking. Refused by the skill; not called a canary.
  • Expand-only migration trap. The state where the expand phase of an expand/contract migration shipped and the contract phase never did, leaving permanent dual-schema liability that compounds across future migrations. The deploy calendar and the in-progress-cycles block of .deploy-ready/STATE.md are the defense.
  • First-deploy blindness. The class of failures that happens only on the first promotion to a new environment: missing env var, unset framework prefix, IAM role that does not exist, .env not read at build time, platform-specific gotchas. Distinct from "works on my machine" because it affects shipping, not development.

What ships

  • SKILL.md with the ready-suite interop standard: eleven frontmatter fields populated, six required sections present. Eleven-step workflow, four completion tiers (Pipelined, Promotable, Reversible, Hardened) totaling 20 requirements, a grep-testable have-nots list, a session state template, and explicit consume/produce contracts with sibling skills.
  • Ten reference files under references/. Load-on-demand table annotates each with the step or tier that loads it.
    • deploy-research.md. Step 0 mode detection (A first deploy, B subsequent, C incident, D pipeline construction, E migration-dominated), destructive-command alert, expand-only migration trap detection procedure.
    • preflight-and-gating.md. 10 pre-flight questions expanded, Mode B subsequent-deploy checklist, the four gate types (build, test, security, approval) with pipeline-enforcement patterns across GitHub Actions, GitLab CI, Argo CD, Flux, Jenkins.
    • deployment-topologies.md. Seven topologies, per-topology first-deploy hazards and rollback characteristics, mixed-topology worked examples (Vercel + Neon, Fly.io, Cloud Run, Lambda + DynamoDB + CloudFront).
    • pipeline-patterns.md. The 8 pipeline gates with good and bad GitHub Actions YAML, same-artifact enforcement via content hashing and registry-tag promotion, the supply-chain pitfalls including pull_request_target fork-RCE.
    • environment-parity.md. The four parity gaps (time, personnel, tooling, fidelity), per-rung parity table, pre-prod parity gap as a named concept, quarterly drift audit.
    • first-deploy-checklist.md. Eleven cold-start gates, per-platform gotchas for Vercel, Netlify, Fly.io, Cloud Run, Lambda, Kubernetes; dry-run rollback procedure.
    • zero-downtime-migrations.md. Expand/migrate/cutover/contract calendar, 10-pattern guardrail catalog with unsafe and safe SQL per pattern, worked 3-deploy column rename across 2 weeks, expand-only migration trap deep dive.
    • rollback-playbook.md. Code-vs-data rollback asymmetry, compensating-forward patterns with worked examples, Knight Capital flag-lineage discipline, destructive-command gate grounded in the Replit 2025 and DataTalks.Club 2025 incidents, incident log template.
    • progressive-delivery.md. Five rollout strategies, paper-canary rule with four required fields, blast-radius rule citing CrowdStrike 2024, Cloudflare 2019, Facebook BGP 2021, readiness probe discipline, preStop graceful-shutdown pattern.
    • secrets-injection.md. Per-topology injection patterns, Docker layer leak class (Truffle Security, GitGuardian, Intrinsec citations), pull_request_target surface, build-time vs runtime split, artifact-level audit commands.
  • Research report (references/RESEARCH-2026-04.md, ~5000 words). Named incidents: Knight Capital 2012, GitLab 2017, AWS S3 us-east-1 2017, Cloudflare WAF 2019, Facebook BGP 2021, CrowdStrike Channel File 291 2024, Replit 2025, DataTalks.Club 2025, timescale/pgai 2025, Docker Hub 10k-image secret-leak class. Tool gap analysis across GitHub Actions, GitLab CI, Argo CD, Flux, Argo Rollouts, LaunchDarkly, Vercel, Netlify, Fly.io. Zero-downtime migration literature survey. Naming-lane analysis. DORA 2024 (7.2% stability drop from AI-assisted delivery), Stack Overflow 2024-2025 (trust-gap framing), GitGuardian 2026 (2x AI-commit secret-leak rate) quantitative framing.
  • SUITE.md at repo root listing deploy-ready at 1.0.0 alongside production-ready 2.5.2 and stack-ready 1.1.1 (sibling copies bumped as patches on this release).
  • README.md with install paths for Claude Code, Codex, Cursor, Windsurf; the "what this skill prevents" incident-to-enforcement mapping across 12 incidents and findings; reference-library index; named-terms section.

Refinements from the dogfood walk

A pre-release paper walk against a realistic solo-dev Fly.io deploy scenario (Node API with a deleted_at column add) surfaced four rough edges. All addressed in v1.0.0:

  • Same-artifact promotion scope clarified. The invariant applies to logical environments (dev, staging, canary, prod). Platform-native region replication and per-region image rebuilds (Fly.io, Cloud Run, Vercel edge) are not drift if the source commit and build configuration are pinned; the artifact-path note records the pin.
  • Compact ladder profile named. The skill does not force a staging rung where one does not exist. A dev-to-prod or dev-to-preview-to-prod ladder is declared as a "compact ladder" and the parity compensations are documented in Step 3. What is forced is that the parity gap is visible, not that a particular rung exists.
  • Solo-dev approval exception. On single-maintainer projects, the approval gate is a distinct second action (signed tag push, deploy command invocation, deploy-marker commit merge) separate from the build step. A pipeline that auto-deploys on push to main with no distinct second action fails the gate even solo, because the whole point of the gate is that shipping is a choice.
  • Expand-only-by-design recognized. Some changes are legitimately expand-only (permanent nullable columns, never-removed enum values, coexisting-forever tables). The calendar records "expand: v1.x. contract: none, by design. reason: ." An absent contract phase is a trap only when deferred and forgotten, not when designed out.
  • Blast-radius exemption for all-at-once prod. All-at-once is acceptable for low-traffic prod services with a named blast-radius justification (user count under threshold, internal-only audience, solo-maintained). Broad user-facing changes still require a non-uniform strategy; the difference is the plan now has to name which context applies.

Compatibility

  • Claude Code (primary)
  • Codex
  • Cursor
  • Windsurf (manual SKILL.md upload)
  • Any agent with skill loading

Suite siblings at release

  • production-ready 2.5.2 (patch bumped for SUITE.md table update)
  • stack-ready 1.1.1 (patch bumped for SUITE.md table update)
  • repo-ready (live, see its own CHANGELOG)

Planned siblings (not yet released): prd-ready, architecture-ready, roadmap-ready, observe-ready, launch-ready.