till/managarten
1
0
Fork 0
mirror of https://github.com/Memo-2023/mana-monorepo.git synced 2026-05-14 22:41:09 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4
managarten/docs/plans
History
Till JS e5d230e599 feat(agent-loop): M1 — policy gate + reminder channel + parallel reads 
Three Claude-Code-inspired primitives for runPlannerLoop, derived from the
reverse-engineering reports in docs/reports/:

1. **Policy gate** (@mana/tool-registry) — evaluatePolicy() gates every tool
   dispatch: denies admin-scope, denies destructive tools not in the user's
   opt-in list, rate-limits per tool (30/60s default), flags prompt-injection
   markers in freetext without blocking. Wired into mana-mcp with a
   per-user rolling invocation log and POLICY_MODE env (off|log-only|enforce,
   default log-only). mana-ai uses detectInjectionMarker only — tool dispatch
   there is plan-only, so rate-limit/destructive checks don't apply yet.

2. **Reminder channel** (packages/shared-ai/src/planner/loop.ts) — new
   reminderChannel callback in PlannerLoopInput. Called once per round with
   LoopState snapshot (round, toolCallCount, usage, lastCall); returned
   strings wrap in <reminder> tags and inject as transient system messages
   into THIS LLM request only. Never pushed to messages[] — the Claude-Code
   <system-reminder> pattern that keeps the KV-cache prefix stable.

3. **Parallel reads** (loop.ts) — isParallelSafe predicate enables
   Promise.all dispatch when every tool_call in a round is parallel-safe,
   in batches of PARALLEL_TOOL_BATCH_SIZE=10. Any non-safe call downgrades
   the whole round to sequential. messages[] always appends in source
   order, never completion order, so the debug log stays linear.
   Default-off (undefined predicate) preserves pre-M1 behaviour.

Tests: 21 new in tool-registry (policy), 9 new in shared-ai (5 parallel,
4 reminder). All 74 green, type-check clean across 4 packages.

Design/plan: docs/plans/agent-loop-improvements-m1.md
Reports: docs/reports/claude-code-architecture.md,
         docs/reports/mana-agent-improvements-from-claude-code.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-23 13:56:40 +02:00
..
agent-loop-improvements-m1.md feat(agent-loop): M1 — policy gate + reminder channel + parallel reads 2026-04-23 13:56:40 +02:00
ai-mission-key-grant.md feat(ai): Mission Grant rollout gating — flag, alerts, runbook, user docs 2026-04-15 14:02:47 +02:00
articles-homepage.md feat(articles): M9 workbench homepage — 4-tab shell + QuickAdd + StatsView 2026-04-22 17:50:38 +02:00
articles-module.md feat(articles): M8 highlights view + stats + dashboard widget 2026-04-22 14:12:18 +02:00
broadcast-module.md docs(broadcast): plan newsletter / broadcast module 2026-04-20 19:32:53 +02:00
data-export-v2.md feat(backup): client-driven v2 snapshot export, drop server-side backup 2026-04-22 18:46:29 +02:00
event-discovery.md feat(events): add Event Discovery — Phase 1 + 2 2026-04-18 15:30:46 +02:00
invoices-module.md docs(invoices): mark all milestones done, list Phase-2/3 open items 2026-04-20 18:23:02 +02:00
library-module.md feat(library): M3 — progress tracking (pages, episodes, issues) + restart 2026-04-17 13:17:22 +02:00
mail-module-plan.md docs(mail): add TODO checklist for remaining Phase 1-4 work 2026-04-13 20:46:00 +02:00
mana-mcp-and-personas.md feat(personas): M2.a-c — persona schemas + admin endpoints + seed pipeline 2026-04-23 13:55:14 +02:00
mana-research-service.md feat(research): Phase 3a — 4 sync research agents 2026-04-17 15:06:12 +02:00
me-images-and-reference-generation.md feat(profile,api): meImages foundation for AI reference generation (M1) 2026-04-23 13:50:53 +02:00
multi-agent-workbench.md docs: flesh out Multi-Agent plan shipping log + Team Workbench forward plan 2026-04-15 22:17:04 +02:00
news-research-module.md docs: flesh out Multi-Agent plan shipping log + Team Workbench forward plan 2026-04-15 22:17:04 +02:00
per-space-vs-user-global-tags.md docs(plans): per-Space vs user-global tags — decision deferred 2026-04-22 15:32:15 +02:00
planner-function-calling.md docs(plans): mark shared-llm tool-call integration as deferred 2026-04-20 18:36:07 +02:00
README.md docs: update tool coverage table + server-side research + templates 2026-04-16 12:35:40 +02:00
scene-scope-empty-state.md docs(workbench): plan for scene-scope empty state 2026-04-18 16:23:13 +02:00
shared-space-smoketest.md feat(spaces): end-to-end shared-space sync (membership lookup + plaintext) 2026-04-20 20:46:53 +02:00
social-relay-module.md feat(spaces): add space types + module allowlist as multi-tenancy foundation 2026-04-20 15:57:57 +02:00
space-scoped-data-model.md docs(plans): point at-rest-sweep row at the restored commit 2026-04-22 22:43:19 +02:00
spaces-foundation.md feat(spaces): migrate calendar module to scoped-db wrapper (pilot) 2026-04-20 16:42:10 +02:00
team-workbench.md docs: flesh out Multi-Agent plan shipping log + Team Workbench forward plan 2026-04-15 22:17:04 +02:00
tipps-module.md Create tipps-module.md 2026-04-22 14:21:57 +02:00
workbench-cards-migration.md docs(plans): revise workbench-cards plan for cards-for-workflows rule 2026-04-22 15:05:48 +02:00
workbench-templates.md feat(templates): generalise to WorkbenchTemplate + ship Calmness pilot (T1) 2026-04-16 01:07:41 +02:00

README.md

Plans

Design + rollout plans, grouped by topic. Plans are long-form docs with baked-in decisions, phasing, open questions, and (when shipped) a history section with commit refs.

AI / Workbench roadmap

The Mana AI Workbench has evolved in three successive planned waves — each one laying foundations the next one relies on:

  User hat einen Companion (v0 — shipped before these docs)
        │
        ▼
  AI Missions + Proposals + Policy + Revert
        │
        ▼
  Mission Key-Grants  ←  ai-mission-key-grant.md  ✅
    (encrypted inputs decryptable by the server runner)
        │
        ▼
  Multi-Agent Workbench  ←  multi-agent-workbench.md  ✅
    (named agents, per-agent policy/memory/budget,
     identity-aware Actor, scene→agent lens)
        │
        ▼
  Team Workbench  ←  team-workbench.md  📝 (not started)
    (multi-user + shared AI context,
     admin lens on team members)
Plan Status Scope
ai-mission-key-grant.md ✅ Shipped Per-mission RSA-wrapped key grant so mana-ai can decrypt allowlisted encrypted records when user opts in.
multi-agent-workbench.md ✅ Shipped Identity-aware Actor + named AI agents owning missions + per-agent policy + scene lens. 28 tools across 11 modules including server-side web-research.
workbench-templates.md ✅ T1 Shipped Generalised templates: 3 agent-templates + 3 non-AI workbench starter-kits. Seed-handler registry for per-module data seeding.
team-workbench.md 📝 Forward-looking TeamSpace with membership, team-encrypted records, admin lens on team members. Reuses Actor.principalId + key-wrapping patterns from the two above.

Cross-references:

Other plans

Plan Topic
mail-module-plan.md Mail module — IMAP/SMTP integration
news-research-module.md News + research pipeline