37 pages tagged
#agent-engineering
agent-computer-interface
concept
ACI: design tools for Agent goals, not API endpoints; when-to-use descriptions; structured errors.
agent-evaluation
concept
Runtime Agent eval: Pass@k vs Pass^k, grader types; eval as the RSI bottleneck.
agent-legibility
concept
Designing systems for the agent's eye, not the human's; logs, metrics, code all agent-readable.
agent-loop
concept
The ~20-line ReAct loop; Workflow vs Agent; five control patterns; loop as optimization target.
agent-memory
concept
Four memory types (working/procedural/episodic/semantic); MEMORY.md + consolidation.
agent-sandboxing
concept
Sandbox as OS-enforced execution boundary; per-OS primitives; Codex-on-Windows design axes.
agentic-context-engineering
concept
ACE: context as evolving playbook; Generator/Reflector/Curator; incremental bullets only.
architectural-invariants
concept
Strict layering enforced by custom linters with embedded fix instructions.
claude-subagents
concept
Spawned Claude instances with isolated context; about isolation, not parallelism.
codebase-as-system-of-record
concept
"If the agent can't find it in the repo, it doesn't exist"; versioned artifacts as truth.
context-engineering
concept
Choosing what enters the LLM context; steering-not-teaching; tiers, Context Rot, compaction.
entropy-and-garbage-collection
concept
Golden principles + scheduled cleanup tasks; debt paid down in small installments.
evolutionary-search-llm
concept
Evolutionary optimization for LLM programs: AlphaEvolve, DGM, Promptbreeder, ADAS, AFlow.
harness
concept
Acceptance + boundary + signal + fallback around the loop; matters more than model choice.
llm-agent
concept
Weng's 2023 anatomy: LLM brain + Planning + Memory + Tool use; ancestor of the wiki's agent pages.
long-running-agents
concept
Initializer + Coding Agent split; state externalized to files; release autonomy in order.
loop-engineering
concept
Anthropic's four-type loop taxonomy (turn/goal/time/proactive); the hand-off ladder.
multi-agent-orchestration
concept
Orchestrator + workers; JSONL inbox protocol, worktree isolation, hallucination amplification.
plan-mode
concept
Read-only exploration mode; refines a plan before any file is touched.
prompt-injection
concept
Source-sink decomposition; tag untrusted content; independent LLM verifier; sandbox as sink-cutting.
ralph-wiggum-loop
concept
Agent self-review pattern: loop until every reviewer + verifier is satisfied; Codex's PR cycle.
react
concept
Reasoning + Acting; the bridge that lets language priors generalize across RL environments.
recursive-self-improvement
concept
RSI: AI improving its own machinery; the near-term path is harness optimization, not weights.
self-improving-harness
concept
An LLM optimizing its own harness: STOP, Self-Harness, DGM; propose-evaluate-accept loops.
self-reflection
concept
Reflexion / Chain of Hindsight / Algorithm Distillation; ancestor of verifier + ralph-wiggum loops.
six-layer-agent-architecture
concept
Tw93's Claude Code frame: CLAUDE.md / Tools+MCP / Skills / Hooks / Subagents / Verifiers.
task-decomposition
concept
Planning by subgoals: CoT, Tree of Thoughts, LLM+P; the decomposition half of agent Planning.
tool-use
concept
MRKL/Toolformer/HuggingGPT/API-Bank; "when & how to call" is the crux; ancestor of ACI + MCP.
verifier-loop
concept
The closing layer; "Claude said done" isn't done — bind acceptance criteria up front.
openclaw
entity
Peter Steinberger's self-hosted personal-AI-assistant daemon; case study in Tw93's Agent essay.
peter-steinberger
entity
PSPDFKit founder; creator of [[openclaw]].
ryan-lopopolo
entity
OpenAI engineer; author of [[2026-04-27-harness-engineering-codex-agent-first]].
tw93
entity
Alibaba/Fliggy engineer; author of all three seed essays and of [[kaku]].
harness-why-it-matters-now
note
Why "Harness" clustered into a thesis in Q1–Q2 2026; five enabling conditions; two postscripts.
2026-04-27-agent-principles-architecture-engineering
source
Tw93's general-Agent companion essay: Harness > model, ACI tool design, OpenClaw walkthrough.
2026-04-27-claude-code-architecture-governance-engineering
source
Tw93's six-layer frame for Claude Code: context, tools, Skills, Hooks, Subagents, verifiers.
2026-07-07-harness-engineering-self-improvement
source
Lilian Weng's Jul 2026 survey: harness engineering as the near-term path to RSI.
esc