Reviews Directory

Type: types/generated-index.md

← Parent

  • Academic Research Skills (note) - Academic Research Skills as a prompt-defined Claude Code research pipeline with narrow executable checks, host-dependent orchestration, protocol-only resume, and conflicting terminal gate rules
  • Agno AgentOS (note) - Whole-system analysis of Agno AgentOS as an open-source execution and control plane, distinguishing its runtime loops from the companion coding-agent and Studio builder loops
  • AI Agents in Depth (note) - Whole-book comparison of AI Agents in Depth with Commonplace, separating broad architectural convergence from differences in memory admission, epistemic warrant, governance, and orchestration
  • AIDE2 (note) - AIDE2's nested harness-rewrite search, private grading and evolved context mechanisms, with documentary limits.
  • Apache Maka (note) - Apache Maka's hosted execution, durable continuation, separate memory acquisition and evaluation authority at a pinned source boundary
  • AREX-Skill (note) - AREX-Skill's repository-skill construction and deployment subsystem: model-directed verification, transactional admission, selective reads and evidence limits
  • arsumbris (note) - arsumbris release-wide review: typed-file memory, adapter-specific governance and instructed knowledge/improvement workflows
  • Autogenesis (note) - Autogenesis as a code-grounded self-evolving agent framework: protocol resources, orchestration, versioned mutation, rollback, and the gaps between its paper artifact and current rewrite
  • beads_rust (note) - beads_rust as a local active-work and coordination substrate: transactional CLI claims and workflow gates, explicit external execution and Git boundaries, and weaker parity across MCP, inherited-context, and shipped instruction paths
  • Claude Code dynamic workflows (note) - How Claude Code's dynamic-workflows API works — a model-authored JS orchestrator over sub-agents — mapped onto the bounded-context orchestration model: what of the tool loop the harness exposes, to whom, and what it withholds
  • Compound Engineering plugin (note) - Compound Engineering's compounding claim separated into product change, project-knowledge retention, and the narrower reflective pathway that can revise project operating instructions but not the installed harness itself
  • ContextPilot (note) - ContextPilot combines task-local memory and context editing with outcome-trained parameters; its control paths and recovery choices bound claims about autonomous context management.
  • DualGraph (note) - DualGraph research workflow: graph-guided search and per-report context reuse, with default-path defects and bounded factual warrant.
  • EAL-bench (note) - EAL-bench's frozen authorization-memory writer/executor workflow, checkpoint controls and limits of its causal-attribution protocol
  • Ecdysis (note) - Ecdysis failure diagnosis, strict-score harness selection, checkpoint reuse, and the external adapter contracts that bound them.
  • Eigenius (note) - Code-grounded review of Eigenius's typed execution substrate and manually loaded host-agent reasoning protocol, with their distinct control and verification limits.
  • Enoch (note) - Enoch's persistent personal-agent core: task recovery, conditional memory read-back, rationale-bearing evolution, and bounded code-adoption gates
  • EvoOntology (note) - EvoOntology's semantic memory and host-led evolution, with supplied-score admission, direct mutation alternatives and content-only version recovery
  • Exo (note) - Exo as a running reflective self-improvement harness: a protected Rust substrate under a fully rewritable executor, allowlisted host control, and a rewind that preserves the record of what was tried
  • Fractal (note) - Fractal as a code-grounded RLM harness: PredictRLM workspace turns, SBX sandbox mounts, headless delegation, and session continuity outside the repo.
  • fragility-grid collection and resume subsystem (note) - fragility-grid evaluates fixed items across prompt and scoring configurations; retained files support analysis and resume, with provenance and completion limits
  • GBrain (note) - GBrain adds provenance-carrying memory, deterministic context push, nightly maintenance and hash-guarded skill delivery to existing agent harnesses; its fact curation catches near duplicates, not contradictions.
  • JEPA-Anything (note) - LLM-hosted world-model design compilation with deterministic consistency checks and generated evidence contracts, distinct from empirical model validation
  • LHTB bundled continuation mechanism (note) - LHTB bundled continuation preserves work and feeds verifier outcomes into later attempts, with narrower feedback and isolation guarantees than its documentation
  • mem: sourced facts and host context (note) - mem extracts sourced facts into Git and supplies host context, with separate publication, recall and withdrawal guarantees
  • MerchantBench: ReAct and optional persistent notes (note) - MerchantBench ReAct baseline combines action feedback and optional persistent notes, with context trimming and source/default limits
  • Meta^n (note) - Meta^n evolves executable solver layers from task traces, retaining code, rationale and task winners; source-level feedback wiring does not establish improved generalization.
  • ModularRSI (note) - ModularRSI's proposal backlog, gated module evolution, dynamic composition, and the limits of its runtime and learning guarantees.
  • oh-my-pi (note) - oh-my-pi's coding loop, effect controls, four memory backends and experiment admission at a pinned implementation boundary
  • OpenRSI: OpenMLE search, experience and training (note) - OpenRSI's released OpenMLE stack: executable program search, retained experience and training interfaces, with evaluation and autonomy boundaries
  • OpenViking session and user memory (note) - OpenViking session and user-memory subsystem: queued extraction, cumulative continuation and bounded recall with explicit host/training exclusions
  • pi-posthorse context-window extension (note) - Pi Posthorse keeps recoverable notes and history while its external Pi host commits fresh context windows
  • Pond (note) - Whole-system analysis of Pond as a durable cross-client session archive, separating its canonical ingest, retrieval, restore, and host policy boundaries
  • Prime Agent (note) - Prime Agent's persistent Python runtime, recursive child sessions, and supplemental-harness refinement, with separate admission and improvement limits
  • PrimeScientist (note) - PrimeScientist's research-plan tree, retained rationale and diagnostics, with score-dependent inheritance and bounded budget/evaluator guarantees
  • Prove2Me (note) - Prove2Me host integration: formal proof and translation review contracts, retained feedback, and shipped Lean extraction helpers.
  • Reflexion HotPotQA reasoning agents (note) - Reflexion HotPotQA reasoning agents: automatic failure-derived prompt memory, exact-match retry control and limits of retained outcome evidence
  • RSIAgent: verified experience, Actor-owned memory and curriculum search (note) - RSIAgent separates task verification, Actor-owned memory reconciliation and curriculum selection; source wiring supports reusable experience but reported gains do not isolate criticism-driven improvement.
  • Semantic Engine as ingest infrastructure (note) - Semantic Engine as code-grounded ingest infrastructure: local SQLite datasets, source chunking, embeddings, query, and visualization surfaces useful before KB promotion.
  • SkillLift (note) - SkillLift's flagship portfolio search: rubric-guided edits, benchmark promotion, retained task memory and a refined-winner recovery limitation
  • SoL-Pi (note) - SoL-Pi wraps Pi with optional fused actions, recoverable observations, checked diagnostic excerpts and plan-driven native compaction
  • Supermemory: the Vercel memory wrapper (note) - Supermemory Vercel wrapper injects retrieved context and uploads conversations, with best-effort saving and an uninspected remote learning bridge
  • Swamp (note) - Swamp as an agent-facing automation control plane: typed resource models, declarative DAGs, remote workers, policy gates, and extension distribution.
  • SwarmWorld (note) - SwarmWorld's simulator-bound agent society: explicit episode memory, executable artifact inheritance, measurement-based skill status and limits on replay and scientific warrant.
  • Tardigrade inference and compaction (note) - Tardigrade derives model progression from event history and separates compaction and schema repair from host durability
  • WikiSkill (note) - WikiSkill's persistent wiki and reversible skill updates, with paper-only evidence and performance-gate limits.
  • WikiSkill (Stahl-G) (note) - Stahl-G's independent WikiSkill implementation: host-agent learning requests, persistent Wiki updates, and score-gated skill adoption.