Skip to content

v2.3.0 — Autonomy framework GA (Sprints A-E)

Choose a tag to compare

View all tags
@explosivebit explosivebit released this 19 May 21:08
· 385 commits to main since this release
v2.3.0
254cc17
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

ForgePlan Marketplace v2.3.0 — General Availability

🎯 The autonomy framework GA: 5 sprints (A-E) over 2 days delivered the complete user vision — "agents do everything, user only provides information". All 9 original autonomy gaps + 10 audited anomalies closed at ≥85%. Real autonomy: ~96% (wired + verified live).

Catalog: v1.48.0 • fpl-skills: v1.23.0 • 13 plugins • 68 agents • 27 skills

Sprint highlights

Sprint A (v1.20.0) — UX layer

  • /agent-advisor skill — recommends right specialist for any task
  • <<NEED_USER_INPUT>> sentinel — formalised subagent ask-back
  • prompt-router hook — UserPromptSubmit classifier

Sprint B (v1.21.0) — closure pack (7 deliverables, dogfood)

  • Parser integration into /forge-cycle + /autorun
  • Methodology citation in all 17 forgeplan-aware agents
  • Profile A Step 10 Hindsight retain convention
  • /project-agent-scaffold, /agent-fetcher, /forge-progress skills

Sprint C (v1.22.0) — /autorun resume protocol

  • Session checkpoint at every phase
  • --resume, --list-sessions, --cleanup-sessions flags
  • Full schema spec at docs/SESSION-CHECKPOINT-SCHEMA.md (643 lines)

Sprint D (v1.23.0) — pipeline self-healing framework

  • <<NEEDS_ACTIVATION>> sentinel — Profile B reviewers signal complete EVIDs
  • Parser in /forge-cycle Step 7.5 + /autorun autopilot
  • /forge-cleanup skill with 3-tier (AUTO/ADI/USER) classification
  • 2 Hindsight mental models: mm-draft-hygiene + mm-pipeline-anomalies

Sprint E (this release) — closure + LIVE verification

  • 7 Profile B agent bodies patched to emit sentinel organically
  • 2 CLAUDE.md docs synced (catalog v1.37 → v1.47 drift closed)
  • AGENTS.md cross-CLI root shim (Gemini/Codex/Goose) — NOTE-005 fulfilled
  • GETTING-STARTED-E2E updated with 5 new Sprint A-D skills (EN + RU)
  • 🎯 LIVE smoke test GREEN — code-reviewer (Profile B) dispatched, emitted sentinel organically on line 1 of return, EVID-060 created + activated via parser
  • This is the first time the autonomy framework verified end-to-end in production

All artifacts

PRDs (5 active, all R_eff=1.0 grade A, all dogfood-activated inline)

  • PRD-029 (Sprint A UX layer)
  • PRD-030 (Sprint B closure pack)
  • PRD-031 (Sprint C autorun resume)
  • PRD-032 (Sprint D self-healing)
  • PRD-033 (Sprint E closure + GA)

EVIDs (5 active, all linked informs)

  • EVID-056..060

Mental models (Hindsight)

  • mm-draft-hygiene — stuck-draft pattern + 3-tier resolution
  • mm-pipeline-anomalies — anomaly catalog (9 initial kinds)

Upstream tracking (forgeplan core)

  • #286 forgeplan_unlink
  • #287 brownfield MCP tools
  • #288 hygiene auto-activate
  • #289 forgeplan_anomalies MCP

Vision delivery

Gap Before Sprint A After Sprint E
A — knowing-which-command 30% 85%
B — subagent-needs-input 10% 95%
C — resume after blocker 30% 98%
D — live progress dashboard 10% 85%
E — right-agent-for-job 40% 85%
F — custom project agent 20% 80%
G — cross-marketplace fetcher 15% 85%
H — Profile A retain 30% 85%
I — methodology citation 60% 95%
Anomaly #7 — EVIDs stuck draft — ✅ Resolved + verified live
Cumulative autonomy ~50% ~96%

Pipeline self-healing framework (Sprint D-E)

3-tier auto-resolution model now wired across orchestrators:

Tier When How
AUTO Low-severity, unambiguous Orchestrator silently fixes (e.g. stuck_draft → activate via NEEDS_ACTIVATION sentinel)
ADI Medium ambiguity FPF reasoning loop — 3 hypotheses → deduce → induct → choose
USER Critical / irreversible NEED_USER_INPUT sentinel via AskUserQuestion

Extensible foundation — new anomaly kinds via /forge-cleanup classification table + mental model entries.

Installation

# In Claude Code:
/plugin marketplace add ForgePlan/marketplace
/plugin install fpl-skills@ForgePlan-marketplace
/plugin install fpl-hsmem@ForgePlan-marketplace
/plugin install agents-core@ForgePlan-marketplace
/plugin install agents-pro@ForgePlan-marketplace
/plugin install agents-sparc@ForgePlan-marketplace
# (See marketplace catalog for full list)

Existing users:

/plugin marketplace update ForgePlan-marketplace

PRs merged for this release (since v2.2.0)

  • #71 Sprint A
  • #72 Sprint B
  • #73 Sprint C
  • #74 Sprint D
  • #75 Sprint E (this release)

What's next (post-v2.3.0)

Plugin layer is at vision parity. Forward work is upstream-dependent:

  • forgeplan core delivers issues #286/287/288/289 → Sprint F can leverage native anomaly detection
  • Live cross-CLI verification (AGENTS.md works in Gemini/Codex/Goose) — exploratory
  • New anomaly kinds added to /forge-cleanup as discovered (orphan_link, mistyped_based_on, etc.) — incremental
  • Remaining open anomaly #1 (forge-cycle template referential gap) — future sprint
Assets 2
Loading