These are systems for operating AI-assisted development—not a prompt bundle. Each makes a different part of engineering work inspectable, recoverable & useful across sessions.
Membrane
Context infrastructure that gathers scoped evidence, ranks freshness and authority, fits it into one budget, then returns a context packet with a receipt for what was admitted, omitted, stale or unavailable. MemRight keeps durable memory local in SQLite.
Context with a receipt
Tether
An evidence ledger for engineering claims. It binds a claim to source hashes, checks & policy; required proof must be present before signoff. When inputs change, dependency-aware invalidation makes old proof fail instead of silently passing.
Proof before signoff
Roundtable
A local hub for steering live Codex & Claude work between seats. Its SQLite WAL store records messages, approvals, handoffs, leases, delivery & deduplication, while session bridges keep provider credentials and hidden context outside the hub.
Live work, coordinated locally
Blueprint
A local SQLite evidence graph for repositories. It maps symbols, files, tests, configuration, plans & ADRs together, then compares documentation claims with current code so a team can find what is true, stale, contradictory or unknown.
Code + document truth
Adapt
A review-gated learning layer that turns accepted corrections into durable preferences. It distinguishes standing rules from one-off facts, keeps an audit receipt, supports rollback & stores accepted records through MemRight’s local SQLite memory.
Corrections that stick
ClaudeCodeX
Run a second isolated Claude or Codex session beside your subscribed desktop app. Your primary session stays untouched; a local proxy routes separate model slots, credentials & profile state for the second instance.
Explore ClaudeCodeX ↗