Fleet-Wide Cross-Repo Learning Synchronization Engine

[github-actions]

Strategic Improvement

Discovery system exists but no fleet-wide knowledge sharing. When one repo learns a failure pattern or successful fix, broadcast it to all fleet repos via a shared memory store. Enables multi-repo deployments to benefit from each other's learnings instantly. A repo that has never seen "ESM import error" learns the fix the moment any other fleet repo encounters and solves it.

Acceptance Criteria

• Shared memory store at ~/.shipwright/fleet-memory/ with per-pattern files
• When sw-memory.sh captures a pattern, broadcast to fleet-memory if in fleet mode
• On pipeline start, load fleet-memory patterns in addition to local memory
• Pattern deduplication: identical patterns from multiple repos merged with vote count
• Pattern expiry: patterns with no successful applications in 30 days archived
• Fleet-wide pattern dashboard showing top-10 learned patterns and adoption rate
• Measure cross-repo pattern reuse rate (target: >20% of patterns used by 2+ repos)

Context

• Priority: P2
• Complexity: full
• Generated by: Strategic Intelligence Agent
• Strategy alignment: P2: Intelligence & Learning

[github-actions]

⚓ Shipwright Pipeline Starting

Field	Value
Template	standard
Complexity	full
Max Iterations	30
Triage	Fleet-wide learning sync spans memory, discovery, fleet, and dashboard subsystems with deduplication, expiry logic, and metrics—a strategic P2 improvement requiring multi-stage validation and testing.
Runner	ubuntu-latest
Risk Level	unknown
Workflow	24282382425

The autonomous pipeline is processing this issue. Updates will be posted as stages complete.

[github-actions]

[github-actions]

[github-actions]

[github-actions]

📋 Implementation Plan

Click to expand full plan (14 lines)

The implementation plan has been written to .claude/plan.md. Here's a summary of the approach:

Architecture: New ~/.shipwright/fleet-memory/ directory with per-pattern JSON files (content-hash named). This minimizes blast radius — existing memory flows are augmented, not replaced.

5 files touched: 2 new (sw-fleet-memory.sh, sw-fleet-memory-test.sh), 3 modified (sw-memory.sh, sw CLI router, package.json)

Key design decisions:

• Per-pattern files instead of single JSON (concurrent-write safe, natural dedup)
• Fleet mode auto-detected via fleet-config.json or $FLEET_MODE env var
• Broadcast is fire-and-forget (silent failure) — non-critical path
• 30-day expiry with 500-pattern cap prevents disk bloat
• Fleet injection is additive (appended after local memory section)

10 tasks covering core engine, memory integration hooks, CLI registration, 16+ tests, and regression verification.

---

Generated by shipwright pipeline at 2026-04-11T12:37:25Z

[github-actions]

[github-actions]

📐 Architecture Decision Record

Click to expand ADR (178 lines)

The plan file is restricted. The ADR content is ready — here it is:

Design: Fleet-Wide Cross-Repo Learning Synchronization Engine

Context

Shipwright's memory system (scripts/sw-memory.sh) captures failure patterns per-repository at ~/.shipwright/memory/<repo-hash>/failures.json. When fleet mode operates across multiple repos (via sw-fleet.sh), each repo learns independently. A build failure solved in repo A is unknown to repo B, so identical failures repeat across the fleet. The existing sw-discovery.sh handles cross-pipeline learning within a single repo (concurrent pipelines sharing discoveries via ~/.shipwright/discoveries.jsonl), but nothing bridges the cross-repo gap.

Constraints:

• All scripts are bash, Bash 3.2 compatible (set -euo pipefail, no associative arrays, no readarray)
• Atomic file writes required (tmp + mv) — multiple daemons may write concurrently from different repos
• Fleet mode is optional — single-repo users must see zero behavioral change
• Memory injection adds context tokens to agent prompts — must be bounded (top-N patterns, ~500 tokens)
• memory_capture_failure() (line 367) and memory_inject_context() (line 977) in sw-memory.sh are the two integration seams

Decision

Architecture: Per-Pattern File Store at ~/.shipwright/fleet-memory/

Each captured failure pattern is stored as an individual JSON file named by the first 16 characters of the pattern's SHA-256 hash. This was chosen over extending global.json or using SQLite because it provides natural deduplication via the filesystem, concurrent-write safety (each pattern is an independent file), and simple expiry (mv to archived/).

Component Diagram

┌─────────────────────────────────────────────────────────────┐
│                      CLI Router (sw)                         │
│  routes `fleet-memory` → sw-fleet-memory.sh                  │
└──────────────────────────┬──────────────────────────────────┘
                           │
┌──────────────────────────▼──────────────────────────────────┐
│              sw-fleet-memory.sh (NEW)                        │
│  ┌──────────────┐ ┌────────────┐ ┌───────────────────────┐  │
│  │ Broadcast     │ │ Load/Rank  │ │ Expire/Archive        │  │
│  │ (write)       │ │ (read)     │ │ (cleanup)             │  │
│  └──────┬───────┘ └─────┬──────┘ └──────────┬────────────┘  │
│         │               │                   │                │
│  ┌──────▼───────────────▼───────────────────▼────────────┐  │
│  │  ~/.shipwright/fleet-memory/                           │  │
│  │  ├── <hash1>.json    (active pattern)                 │  │
│  │  ├── <hash2>.json    (active pattern)                 │  │
│  │  └── archived/                                        │  │
│  │      └── <hash3>.json (expired)                       │  │
│  └───────────────────────────────────────────────────────┘  │
│  ┌──────────────┐ ┌────────────┐                            │
│  │ Dashboard     │ │ Status     │ (CLI subcommands)         │
│  └──────────────┘ └────────────┘                            │
└─────────────────────────────────────────────────────────────┘
                    ▲                    │
                    │ source + call      │ returns markdown
                    │                    ▼
┌─────────────────────────────────────────────────────────────┐
│              sw-memory.sh (MODIFIED)                         │
│                                                              │
│  memory_capture_failure()  :440                              │
│    └── if _is_fleet_mode() → fleet_memory_broadcast()       │
│                                                              │
│  memory_inject_context()   :977                              │
│    └── if _is_fleet_mode() → fleet_memory_load()            │
│        └── appends "## Fleet-Wide Learned Patterns" section  │
│                                                              │

</details>

---
_Generated by `shipwright pipeline` design stage at 2026-04-11T12:42:54Z_

[github-actions]

[github-actions]

🔨 Build started — shipwright loop with 30 max iterations, 1 agent(s), model: opus

[github-actions]

[github-actions]

❌ Pipeline failed at stage build after 10m 54s.

No log available