Refactor HydrationConfig to stacked HydrationSource list
Wire knowledge graph traversal into passive hydration
Wire knowledge graph traversal into passive hydration
Prevent persistent agent 'promise loop' (says it will work but does not execute tools)
Make omni .envrc idempotent/non-noisy for non-interactive direnv exec
agentd top: j/k navigation, ^C watch exit, Activity cache refresh
Refactor shared systemd helpers across agentd and deployer
Fix typecheck.sh jq parsing for LocalRun rundeps
sd: investigate low acceptance rate + verify idle A100s shut down
Intent: Impl-vs-spec verification
Persistent agent env drift can set AGENTD_AGENT_COMMAND=agent and break services
Fix ambiguous Data.ByteString.Base64 import in Omni/Health/Intervals
Design and implement mkSkill Nix helper + skill CLI
Align Anthropic OAuth flow with pi-mono
Notify owner on LLM provider failover
workflows: add system_prompt to reviewer + integrator frontmatter; drop obsolete coder/overseer
Pipeline scheduler: verify commits reach live before marking status=done (fixes orphan-commit bug)
Port pi's system prompt to Omni/Agent coding agents
Design compaction strategy for Omni/Agent coding agent
Fix Omni/Ipt.py build failure
Research: unify trace compatibility for persistent sessions
Skill drift: ui-preview example references removed agentd-studio path
Cleanup: remove legacy in-process agent code from Omni.Agentd.Daemon
Bug: task CLI fails to read mixed-case enum values in tasks.db
Bug: agentd service restart can hang in stop-sigterm for ~2m
Intent Phase 4 follow-up: fix exec_cmd code type (Nat→Int)
Intent Phase 4 follow-up: align :foreign ordering semantics (parser/docs/tests)
Intent Phase 4 follow-up: include declForeign in hash/store persistence
Intent Phase 4 follow-up: make :foreign python executable in codegen
Intent: include regression cases in default test workflow
Intent: resolve remaining LANG.md contract contradictions
Intent: make query-holds KB scoping strict by default
Phase 4: Health metrics at /ava/health
Phase 3: Engagement audit at /ava/engagement
Phase 3: Context inspector
Phase 2: Memory browser at /ava/memories
Phase 2: Message detail view at /ava/chats/:chatId/messages/:msgId
Ava Audit Dashboard
Quant Phase 2: Alpha + Portfolio + Web
Quant Phase 1: Data + Signals + Eval
Secrets not provisioned after reboot
Web architecture improvements (epic)
Set up SMTP secrets for fund-digest service
Migrate fund queries from hledger CLI to hledger-lib Haskell API
Agentd daemon pi.log writing is broken - handle closed immediately
Send to Review: integrate agent work into live branch
Worktree diff view and file browser
Bug: ship/push may report manifest update without persisting
Add torch, mamba-ssm, causal-conv1d to omnirepo Python nix deps
Guard against web deploy rollback from divergent worktrees
MVP 3: Diverse model review (devil's advocate)
MVP 2: Single-model auto-spec refinement loop
MVP 1: Executor GO/NO-GO gate
MVP 0: Baseline metrics for task pipeline
Epic: Automated Task Up-Speccing Pipeline
JIT context rehydration via request_context tool
MVP 0: Baseline metrics instrumentation
General OAuth 2.0 integration for Telegram bot
Agent auth path drift: pipeline/agentd uses ~/.local/share/agent but fresh creds are in ~/.pi/agent
Agentd: avoid zombie 'running' status when container exits before terminal event
Pipeline: don't move failed runs to Review
Investigate mbsync duplicate UID errors in Maildir
Task system: validate namespace exists in repo when setting it
Broadcom (AVGO) 10-K deep dive
Research: SRAM companies 10-K deep dives — post-Groq/NVDA inference thesis
Research: Data center cooling companies 10-K deep dives
Research: Arista Networks (ANET) 10-K deep dive
Newsreader: breadcrumbs reset when viewing file content
Support ~/work as a second Pipeline domain
Fix agentd buffet provider support (claude-code, codex)
fund command - domain budget management
DevBrowser fails without local Playwright browser binaries
Benchmark memory system against LoCoMo eval
Interactive step-through debugger
Socket message protocol and handler
Ava dual-channel listener (Telegram + socket)
Pi extension for agent completion notifications
Add user_profiles table for per-user preferences
Pre-compaction memory flush
Graph: Terminal UI for navigation and composition
Graph: TUI event handling
Graph: TUI rendering
Graph: TUI state and navigation
Agent Graph (epic)
Enforce PromptIR token budgets in Sequential
Wire AdaptiveContextConfig thresholds
Review memory improvements and prompt IR
Research: auto-detection for best-of-N parallel execution
Integrate weeder for Haskell dead code detection
Prototype: LLM as universal semantic interpreter (temperature=0)
Design: leverage pure cognitive compute in Bayesian prompt framework
Research: rate-distortion theory for prompt compression
Research: Bayesian calculus for prompt composition
Research: static prompt analysis using open model internals
Add capability-based permissions to agentd workflows
Drip scheduler for follow-up campaigns