Milestone: v1.0-GA — Production-Credible Release

Target Maturity: ga Status: ACTIVE Started: 2026-04-30 Target Completion: ~2026-07-18 (8–10 week scope, comfortable pace per D7) Driving Release: v1.0.0

Goal

Earn ADD the right to call v1.0 "GA." That means closing the four hard blockers Swarm 1 surfaced, shipping the engineering work named in the reworded next_promotion_criteria, repositioning the methodology to survive host-runtime feature absorption (the strategic risk Swarm 3 named), and proving multi-runtime install reproduction in CI.

This milestone is the credibility cycle, not the feature cycle. The methodology surface is broad enough; v1.0's job is making the public claim surface match reality and proving the install path actually works.

Driving Context

Three parallel research swarms (current state, unimplemented specs/plans inventory, market review) returned 1,962 lines of analysis. The synthesis lives at docs/v1.0-roadmap.md. Headline findings:

~1.5 of 6 published GA criteria met as of v0.9.4. Telemetry contract present but emission unverified; Codex hook stderr never reaches the agent; AGENTS.md regen ritual missed since v0.9.1; marketplace install has zero CI smoke.
9 specs carry stale Draft/Implementing frontmatter despite being operationally Complete. Documentation drift, not unimplemented work.
9 F-IDs from plugin-family-release-hardening.md are unshipped. F-006/F-007 architectural; F-008/F-009/F-010 Codex packaging behind Q-001; F-012 hook feedback gates 2 downstream blockers; F-015/F-016/F-019/F-020 are smaller pieces of catalog/install/site coherence.
The market shifted faster than ADD shipped. Cloud agents (Claude Code Routines), visual multi-agent UX, benchmarked memory, skill task-evals + security scoring, the spec-driven branding war (Spec Kit + DeepLearning.AI canonical) — all 2026 developments where ADD is invisible or behind.
ADD's only durable moat is maturity-aware methodology. The poc → alpha → beta → ga ladder maps to the trust-gradient problem the Anthropic 2026 Trends Report names as THE 2026 issue. Most other ADD differentiators (hooks, sub-agent dispatch, AGENTS.md generation) have been absorbed by host runtimes.

The v1.0 path is mostly truth + integration + Codex coherence + repositioning, not new features. The full research is preserved at docs/research/v1.0-swarms/.

Seven decisions (D1–D7) were locked through maintainer review on 2026-04-30. Disposition documented in docs/v1.0-roadmap.md Part 8.

Success Criteria

Per the reworded .add/config.json:next_promotion_criteria (D1):

#1 Guardrail suite running in CI and release-blocking
#2 Real Claude + Codex install smoke in CI
#3 Per-runtime capability matrix in release notes
#4 Beta exercise evidence: ≥3 minor or patch releases past v0.9.0 shipped with no post-tag rollback; ≥1 community PR merged during the beta cycle; AC-025 release-evidence bundle generated for the v1.0 candidate release
#5 Marketplace submission approved
#6 ADD verified in active use beyond the maintainer's solo work (maintainer-attested per D1 reword)

Status today: #1 near-met (running; release-blocking branch protection pending v0.9.6); #4 partially met (3 of 3 release-cycles + community PR met; release-evidence bundle pending); #6 met (maintainer attestation per D1).

Status update 2026-07-18 (v0.10.0 cycle, per specs/install-path-confirmation.md):

Decision D8 — consolidation + submit-before-promote. The planned v0.10.0 and v0.11.0 cycles are consolidated into one v0.10.0 "install path confirmed" RC. Marketplace submission (#5) moves to AFTER the RC's install smoke is green, so the external review exercises a machine-verified artifact; v1.0.0 is reserved as the promotion tag on approval. The token-audit arc (v0.9.8–v0.9.11, unplanned) displaced the original v0.9.6/v0.9.7 deliverable schedule; D5 (methodology.md), F-015/F-016/F-019, /add:post-release, and scripts/install-claude.sh are re-scoped out of the GA gate (none block the six criteria) and move to v1.1 unless capacity allows.
#1 MET: release.sh refuses to tag unless HEAD's CI checks are green (--no-verify-ci emergency override prints loudly; covered by the release-tooling fixture suite), and branch protection on main now requires 23 contexts — both install smokes + every guardrails job (enforce_admins false to preserve the maintainer direct-push flow; see docs/release-signing.md for the re-apply command).
#2 FULLY MET (2026-07-18 PM): both smokes green in CI end-to-end WITH live agent legs — real marketplace install + /add:init on Claude, real curl install + /add-init on Codex 0.144.5, both producing valid .add/config.json. The live runs also surfaced #23 (init --quick not headless-safe) and #24 (Codex ≥0.14x rejects ADD sub-agent TOMLs + hooks.json schema — matrix now says "Broken on ≥0.14x"; fix targets v0.10.1).
#3: docs/capability-matrix.md created; release.sh appends it to every release's notes; SECURITY.md reframed to point at it (AC-027 honest-docs fallback in place regardless of the F-012 spike outcome).
#4: scripts/release-evidence.sh implements the AC-025 bundle including a migration-graph reachability check — which immediately caught a real defect: v0.7.3 had no outgoing hop (stranded users, same class as the v0.8.1→v0.9.3 incident). Fixed by adding a 0.7.3→0.8.0 hop.
Q-MS-003 answered (D6): telemetry is NOT emitting. .add/telemetry/ does not exist on the dogfood project after months of use — the spec exists but no hook writes it. Finding recorded; emission implementation deferred (does not gate the six criteria; PRD metrics that depend on it stay unmeasurable until then).

Plus the cycle's own deliverables (per D2/D3/D5/D6 lock):

Deferred to v1.1 / M4 (out of scope for this milestone — see § Out of Scope):

D2 Tier 3 AC-012 host-neutral methodology + runtime overlay split
D4 ADD as MCP server
Dashboard telemetry panel (D6 Tier 2)
F-006/F-007 full architectural close
Cursor + Cline runtime adapters
M3 deferrals: /add:parallel, /add:eval, /add:cycle rename, brownfield delta-spec mode, governance maturity bands, cross-tool memory schema

Appetite

8–10 weeks of engineering. Per the 2026-06-14 maintainer override to D7, the arbitrary 60-day beta calendar floor is dropped — v1.0 is gated on Anthropic marketplace approval (criterion #5) plus the substantive criteria, not a fixed-day count. The marketplace submission timeline is external and is now the binding constraint: GA tags when approval lands and the substantive gates are green, whenever that is. Sequential engineering, low regression risk during v0.10/v0.11 Codex coherence work.

Features

Hill Chart

                    figuring it out  |  executing
                                     |
  ●  D2 Tier 2 conditional           |
   ●  F-012 spike outcome (D3)       |
                                     |
                                     |  ●  v0.9.5 hygiene + truth pass
                                     |   ●  /add:post-release skill (D2 specced)
                                     |    ●  catalog gen + site sync (F-019)
                                     |     ●  D2 Tier 1 substitution close
                                     |      ●  D5 methodology.md + mirror
                                     |       ●  D6 telemetry verification + Tier 1
                                     |        ● F-015 + F-016
                                     |         ● Codex coherence (F-008/9/10)
                                     |          ● install smoke (AC-022/023)
                                     |
                  uphill              downhill
                  (design)            (implementation)

Per-release feature detail

Release	Sizing	Touches	Independence	Notes
v0.9.5	Tiny (~1 day)	docs / specs / plans / config	Independent	Truth pass; no engineering changes
v0.9.6	Small (~1 wk)	new skill body + CI workflow + new install script	Independent	`/add:post-release` already specced
v0.9.7	Medium (~2 wks)	catalog generator + site sync + `methodology.md` + telemetry hook	Mild deps on v0.9.6 release infrastructure	Methodology + telemetry parallel-safe
v0.10.0	Medium-Large (~3 wks)	config schema + installer + F-012 spike	Independent of v0.9.x; touches `compile.py` (Tier 2 conditional)	F-012 spike outcome decides Tier 2 scope
v0.11.0	Large (~3 wks)	Codex adapter re-baseline + container CI	Touches `runtimes/codex/` heavily	Highest blast radius; comfortable pace warranted
v1.0.0	Small (~1 wk)	release-evidence bundle + Claude install smoke + maturity promotion	Final assembly	External marketplace gate may extend

Cycle Plan

v0.9.5 — Hygiene + Truth Pass (~1 day, target 2026-05-02)

Doc-only release. Removes the smokescreen so v0.9.6+ work happens against accurate documentation.

README counts (rules 11→19, templates, knowledge), roadmap section reframe (drop "v0.4.0 — Next" historic block), methodology-led intro (D5 seed)
CLAUDE.md template/knowledge counts; add core/lib/, core/security/, core/references/ to repository structure
runtimes/claude/CLAUDE.md skill table verification
PRD header date update; § 4 v1.0 reframe (current PRD describes v1.0 surface that's mostly already shipped); § 3 telemetry-claim qualification (D6 Tier 0 prep)
core/rules/maturity-loader.md cascade matrix completed for the 4 rules added in v0.9.0 (cache-discipline, injection-defense, secrets-handling, telemetry)
AGENTS.md regen via python3 scripts/generate-agents-md.py --write
9 spec frontmatter flips (Draft/Implementing → Complete) — see Swarm 2 § A.2
17 plan files: append Status: Complete (vX.Y.Z) headers
Retire specs/timeline-events.md (Supersede with one-line note pointing at telemetry-jsonl + dashboard)
Architect/Editor model-role rule (M3 deferral, one-paragraph addition to core/rules/)
Amend specs/plugin-family-release-hardening.md § 2 finding table: F-006/F-007 target → v1.1.0 with rationale; § 3 retag AC-010/AC-011/AC-012 to v1.1.0
.add/config.json:next_promotion_criteria reworded per D1 ✓ done
CHANGELOG entry summarizing as "v1.0 prep — credibility hygiene"

Validation: python3 scripts/compile.py --check, python3 scripts/validate-frontmatter.py, bash tests/rule-parity/test-rule-parity.sh, all 12 fixture suites green.

v0.9.6 — Post-Release Skill + Release-Blocking Gates (~1 wk, target 2026-05-12)

Operationalize the publication ritual; promote the CI suite from "passing" to "release-blocking."

/add:post-release skill ships per specs/post-release-publication.md (32 ACs, plan exists)
AC-024 — guardrails release-blocking via branch protection on main + release.sh precondition checking guardrails CI green
scripts/install-claude.sh (AC-028) — public-user install bypassing marketplace auth, tag-pinned URL, checksum verification
Migrations CI guard — fail PRs that bump core/VERSION without a migration entry (closes the v0.8.1→v0.9.3 silent-skip pattern that surfaced in v0.9.4)

Validation: /add:post-release v0.9.6 runs against itself dog-foods the new skill; 13-job CI green.

v0.9.7 — Catalog + Site Sync + Telemetry + Methodology (~2 wks, target 2026-05-30)

The largest v0.9.x cycle. Multiple disjoint feature streams that can ship via parallel swarms.

F-019 catalog generator (scripts/generate-command-catalog.py) — single source for skill counts, command syntax, risk levels, dispatch policy, write behavior. Consumed by README inserts, marketplace.json, runtime AGENTS/CLAUDE docs, website inputs.
Site-metrics + skills-page generators — closes release-materials.md open follow-ups; MountainUnicorn/getadd.dev/index.html metrics bar and docs/skills.html regenerated from core/.
/add:announce skill — CHANGELOG → blog post draft. Closes uneven blog cadence.
AC-027 per-runtime capability matrix in release notes template + SECURITY.md reframe (pairs with D3 fallback if F-012 spike fails).
D2 Tier 1 substitution close — add CLAUDE.md, ~/.claude/add, ~/.claude, /add:<name> regex to CODEX_TOOL_SUBSTITUTIONS. Closes ~80% of F-006 visible leak in dist/codex/ output.
D5 deliverable — docs/methodology.md canonical (~3000 words, citable, version-pinnable via GitHub permalink) + getadd.dev/methodology mirror page.
D6 verification step (1 hr) — confirm whether .add/telemetry/{date}.jsonl is actually populated on dog-food. If broken, escalate scope.
D6 Tier 1 — hook-driven emission (runtimes/claude/hooks/emit-telemetry.sh) + scripts/collect-telemetry.py reference local collector + fixture tests + docs/telemetry.md.

v0.10.0 — Config Schema + Installer Ownership + F-012 Spike (~3 wks, target 2026-06-20)

The first release that formalizes contracts. Decision point on F-012 hook-feedback adapter via the D3 spike.

F-015 config schema + migration graph validator (core/schemas/config.schema.json)
F-016 installer ownership manifest (scripts/install-codex.sh writes dist/codex/ownership.json; backups overwritten files; --dry-run; uninstall only owned files)
D3 F-012 spike (2 days time-boxed) — research Codex CLI's UserPromptSubmit hook surface; prototype runtimes/codex/hooks/inject-security-events.sh. If injection works cleanly: ship the adapter (1-2 more days). If brittle: fall back to (b) AC-027 docs already in v0.9.7.
Once F-012 outcome known: enable secrets-scanner hard-block at git push (Claude unconditional; Codex conditional on (a) outcome) and prompt-injection escalation from warn to blocking on critical patterns.
D2 Tier 2 conditional — AC-010 (adapter contracts authoritative; move CODEX_TOOL_SUBSTITUTIONS dict from compile.py constant to adapter.yaml) + AC-011 (introduce ${ADD_HOME}, ${ADD_RUNTIME_ROOT}, ${ADD_USER_LIBRARY} neutral path variables). Land if v0.10 capacity allows; defer to v1.1 otherwise.

v0.11.0 — Codex Coherence + Real Install Smoke (~3 wks, target 2026-07-11)

Single largest GA-credibility hardening cycle.

Q-001 resolved — Codex CLI re-baselined to current stable (Codex shipped v0.140+ since ADD pinned v0.122). Either re-pin or move to "latest minor of Codex CLI tested in CI."
F-008 — Codex marketplace/package format matches pinned CLI's actual conventions
F-009 — Codex skill policy metadata (agents/openai.yaml) shape validated
F-010 — Codex hooks/config enablement (plugin-relative or safe global merge with detection)
AC-035 — containerized Codex CI runner (Docker image with pinned Codex CLI; runs install smoke + /add-init + skill discovery)
AC-023 — real Codex install smoke in CI

v1.0.0 GA — Real Claude Smoke + Release Evidence + Promotion (~1 wk, target 2026-07-18)

The promotion release.

AC-022 real Claude install smoke in CI — containerized Claude Code runner; installs from marketplace; runs /add:init --quick; verifies .add/config.json produced
AC-025 release-evidence artifacts — supported runtime matrix, known limitations, install smoke outputs, command catalog snapshot, version map, migration coverage. Generated alongside the GitHub release.
/add:promote --execute --target ga — runs the cascade gap analysis; cascade matrix gets a "plugin-source mode" addendum so cascade rows that assume a deployment ladder (Tier 2-3, two reviewers, daily standups) are explicitly N/A for plugin projects.
Marketplace submission — submit to official Claude Code marketplace registry (external dependency on Anthropic review).
CHANGELOG [1.0.0] — release notes lead with maturity ladder + multi-runtime + structured methodology (the moats), are honest about what's warn-only on Codex (per AC-027), and ship with measurable release evidence.

Parallelism Analysis

Within v0.9.5

Five disjoint file sets — can be split across 3-5 parallel agent swarms (Wave 1 of M3-style execution):

Swarm	Files	Source
Top-level docs	`README.md`, `CLAUDE.md`, `runtimes/claude/CLAUDE.md`, `CONTRIBUTING.md`	Truth pass
Specs / plans	`specs/.md` (9 frontmatter flips + 1 Supersede), `docs/plans/.md` (17 closeouts), retire `specs/timeline-events.md`	Hygiene
PRD + maturity + retarget	`docs/prd.md`, `core/rules/maturity-loader.md`, `specs/plugin-family-release-hardening.md`	Reframe
Architect/Editor rule	new `core/rules/architect-editor.md` (or addition to existing rule)	Tiny new rule
Final assembly	AGENTS.md regen, migrations.json hop, VERSION/config.json/README badge bump, CHANGELOG, compile, gates, tag	Orchestrator only

Within v0.9.7

Four parallel feature streams:

Stream	Files	Notes
F-019 catalog generator	`scripts/generate-command-catalog.py` (new), README integration	Independent
Site sync	`MountainUnicorn/getadd.dev` index.html metrics + skills.html regen	Different repo; no conflict
`docs/methodology.md` + `getadd.dev/methodology` (D5)	new file in plugin repo + new HTML page in site repo	Cross-repo; small
D6 Tier 1 telemetry	`runtimes/claude/hooks/emit-telemetry.sh` (new), `scripts/collect-telemetry.py` (new), test fixtures	Independent
`/add:announce` skill	`core/skills/announce/SKILL.md` (new), templates	Independent
D2 Tier 1 substitution	`scripts/compile.py` substitutions + test fixture	Mild conflict surface — same file as catalog gen if it touches compile output

Within v0.10.0 / v0.11.0

Sequential by design — Codex coherence work has high blast radius and conflicts with parallel compile.py edits. Comfortable pace.

Risks & Mitigations

Risk	Likelihood	Impact	Mitigation
F-012 spike reveals Codex hook surface is observe-only	Medium	Medium — Codex enforcement stays warn-only at v1.0	D3 fallback: AC-027 honest matrix already shipped in v0.9.7; v1.0 marketing language calibrated
Codex CLI re-baseline (Q-001) takes longer than 1 week	Medium	High — v0.11.0 is binding constraint	Time-box research; if 1-week gate fails, defer hardening details to v1.1 and ship v1.0 with current Codex pin + documented limitation
Marketplace submission rejected or delayed	Medium	High — external gate cannot be accelerated	Submit early in v0.10/v0.11 cycle; iterate on rejection feedback in parallel with v0.11 work
Container CI for Codex/Claude is hard to stand up	Medium	Medium	Use prebuilt Docker images where possible (Anthropic + OpenAI both publish CLI containers); fall back to GitHub Actions hosted runners if blocking
`/add:post-release` reveals v0.9.5 truth pass missed items	Low	Low	First dog-food run is on v0.9.6 itself; corrections feed v0.9.7
D6 verification finds telemetry emission entirely broken	Medium	Medium	Tier 1 hook-driven emission is the fix; scope already accounts for it
F-015 config schema retrofitting breaks existing `.add/config.json` files	Low	Medium	Schema validates as warning at v0.10; promotes to error at v1.0 with migration step
Architect/Editor rule (M3 deferral) reveals deeper governance issue	Low	Low	One-paragraph addition; if it grows, defer to v1.1

Out of Scope (deferred to v1.1 / M4)

These items are explicitly NOT in v1.0 per the locked decisions:

D2 Tier 3 AC-012 host-neutral methodology + runtime overlay split — true architectural cycle work, M4
D4 ADD as MCP server emission target — M4 alongside Cursor + Cline adapters
D6 Tier 2 dashboard telemetry panel — v1.1 enhancement
Cursor + Cline runtime adapters — M4 architecture-validation deliverable
F-006/F-007 full close — partial in v0.9.7/v1.0 (Tier 1 + conditional Tier 2); architectural close in M4
From M3 deferrals: /add:parallel, /add:eval, /add:cycle → /add:arc rename, brownfield delta-spec mode, governance maturity bands, cross-tool memory schema, Routines/Loop integration adapter
Computer use for spec verification — Swarm 3 § A7, COULD priority, defer until beta+ users specifically request
Persistent semantic / temporal memory adapter — Swarm 3 § A6, COULD priority, integration via MCP rather than ADD-built

These remain in specs/plugin-family-release-hardening.md and docs/v1.0-roadmap.md § Part 4 as the M4 scope.

Open Questions

ID	Question	Owner	Target
Q-MS-001	Marketplace submission review timeline — when does Anthropic typically respond? Affects v1.0 calendar but not engineering work.	Maintainer	v0.10
Q-MS-002	F-012 spike outcome — does Codex UserPromptSubmit hook allow context injection, or is it observe-only? Determines whether (a) ships or (b) is permanent.	Engineering	v0.10 (2-day spike)
Q-MS-003	D6 verification result — is telemetry actually emitting on dog-food today? Determines whether D6 scope expands.	Engineering	v0.9.7 (1-hr check)
Q-MS-004	D2 Tier 2 capacity in v0.10 — does the F-012 spike + F-015 + F-016 leave room for the adapter-contract refactor? Determines whether AC-010/AC-011 ship in v1.0 or defer to v1.1.	Engineering	v0.10 mid-cycle
Q-MS-005	Cursor + Cline adapter authorship — solo maintainer or community contribution? Affects M4 milestone sizing.	Maintainer	post-v1.0

Validation Per Cycle and Milestone

Per release (standard ADD gates)

python3 scripts/compile.py --check clean
python3 scripts/validate-frontmatter.py clean
All fixture suites green (tests/* — count grows from 12 to ~15 across v0.9.5 → v1.0.0)
CHANGELOG [X.Y.Z] entry present
migrations.json hop entry added (per the v0.9.4 lesson)
AGENTS.md regenerated
/add:post-release vX.Y.Z walked (from v0.9.6 onward)

Per milestone (v1.0.0)

All 6 reworded next_promotion_criteria items met or maintainer-attested
Release-evidence bundle artifact attached to GitHub release
CHANGELOG [1.0.0] published with full release notes
Maturity promotion beta → ga executed via /add:promote --execute --target ga
Per-runtime capability matrix in release notes (AC-027)
No new F-NNN findings raised post-tag (graceful tolerance: 1-2 minor findings expected per the v0.9.x F-NNN cadence; should not include shipping-blockers)

Closing

ADD at v0.9.4 has the methodology surface. v1.0 is the credibility cycle that earns the GA tag.

Path: 6 releases over 8–10 weeks, comfortable pace, sequential engineering at the high-blast-radius v0.10/v0.11 points. Calendar floor dissolved (D1); engineering + marketplace approval are the binding constraints. v1.0 ships clean, v1.1 / M4 picks up the architectural cycle (host-neutral kernel + Cursor/Cline adapters + MCP server emission) post-GA.

The most important thing this milestone does NOT try to do: ship v1.0 with all loose ends closed. Some loose ends are M4 work; trying to land them in v1.0 displaces the credibility-hardening cycle that v1.0 actually needs. Hold the line on scope.

Companion docs: docs/v1.0-roadmap.md (synthesis), docs/research/v1.0-swarms/ (raw research output), docs/release-materials.md (post-release checklist used per release).

FilesExpand file tree

v1.0-ga.md

Latest commit

History