agentic-harness-patterns

Name: agentic-harness-patterns
Author: keli-wen

165installs48forks

How to install agentic-harness-patterns?

keli-wen/agentic-harness-patterns-skill/agentic-harness-patterns

$npx -y skills add keli-wen/agentic-harness-patterns-skill --skill agentic-harness-patterns

Installs into the current project.

Summary

Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.

Files · 1

View on GitHub

SKILL.md

1	# Agentic Harness Patterns
2
3	Production AI coding agents are not just an LLM calling tools in a loop. The harness — memory, skills, safety, context control, delegation, and extensibility — is what separates a demo from a production system.
4
5	For: Engineers building or extending coding-agent runtimes, custom agents, or advanced multi-agent workflows.
6	Not for: Prompt engineering, model selection, generic software architecture, or LLM API basics.
7
8	All principles are distilled from production runtime decisions. Claude Code is used as grounding evidence, not as the only possible implementation.
9
10	## Choose Your Problem
11
12	\| If you want to... \| Read \|
13	\|---\|---\|
14	\| Make the agent remember and improve over time \| [Memory](#1-memory) \|
15	\| Package reusable workflows and expertise \| [Skills](#2-skills) \|
16	\| Let the agent use tools powerfully but not dangerously \| [Tools and Safety](#3-tools-and-safety) \|
17	\| Give the agent the right context at the right cost \| [Context Engineering](#4-context-engineering) \|
18	\| Split work across multiple agents without losing control \| [Multi-agent Coordination](#5-multi-agent-coordination) \|
19	\| Extend behavior with hooks, background tasks, or startup logic \| [Lifecycle and Extensibility](#6-lifecycle-and-extensibility) \|
20
21	Before you start building: Read the [Gotchas](#gotchas) — these are the non-obvious failure modes that cost the most time.
22
23	---
24
25	## 1. Memory
26
27	User problem: "My agent forgets corrections and project rules between sessions."
28
29	Golden rule: Separate what the agent knows (instruction memory) from what the agent learns (auto-memory) from what the agent extracts (session memory). Each layer has different persistence, trust, and review needs.
30
31	When to use: Any agent that operates across multiple sessions or needs to accumulate project-specific knowledge over time.
32
33	How it works:
34
35	- Instruction memory is curated, hierarchical configuration injected into system context in priority order (org-wide → user → project → local; local wins). This is where project conventions, coding standards, and behavioral rules live. It is human-authored and stable.
36	- Auto-memory is agent-written persistent knowledge with a type taxonomy (user / feedback / project / reference) and a capped index. Saving is two-step: write a topic file, then update the index. The cap prevents unbounded growth — without cleanup, recent entries silently disappear.
37	- Session extraction runs as a background agent at session end. It directly writes to auto-memory — topic file then index — following the same two-step save invariant. A mutual-exclusion guard ensures that if the main agent already wrote memory during the turn, the extractor skips entirely. This is the autonomous learning loop.
38	- Review and promotion audits across all memory layers and proposes cross-layer moves (auto-memory → project conventions, personal instructions, or team memory). It never applies changes autonomously — proposals require explicit user approval.
39
40	Start here: Define your memory layers (instruction, auto, extraction). Implement the two-step save invariant (topic file, then index). Add background extraction only after the core write path is stable.
41
42	> In Claude Code: Use `/remember` to audit and promote auto-memory entries across layers.
43
44	Tradeoffs:
45
46	- More memory layers = richer recall but higher maintenance burden. Without periodic pruning, index caps cause silent data loss.
47	- Session extraction adds latency at session end but dramatically improves cross-session learning.
48
49	Go deeper: [references/memory-persistence-pattern.md](references/memory-persistence-pattern.md)
50
51	---
52
53	## 2. Skills
54
55	User problem: "I want my agent to reuse workflows and domain knowledge without re-explaining them every time."
56
57	Golden rule: Skills are lazy-loaded instruction sets, not eagerly injected prompts. Discovery must be cheap (metadata only); the full body loads only on activation.
58
59	When to use: Any agent that needs reusable, composable workflows activating on matching user intent.
60
61	How it works:
62
63	- Discovery is budget-constrained: the agent sees a compact listing of all available skills (name, description, and when-to-use hint concatenated per entry), each hard-capped at a fixed character limit, with the total capped at roughly 1% of the context window. Front-load your trigger language — tails get truncated.
64	- Loading is lazy: only metadata enters the always-on context. The full skill body loads only when the skill activates, k

People who liked this also viewed

6 picks

Type

agentic-harness-patterns

165installs48forks

View on GitHub

How to install agentic-harness-patterns?

keli-wen/agentic-harness-patterns-skill/agentic-harness-patterns

$npx -y skills add keli-wen/agentic-harness-patterns-skill --skill agentic-harness-patterns

Installs into the current project.

Summary

Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.

Files · 1

View on GitHub

SKILL.md

1	# Agentic Harness Patterns
2
3	Production AI coding agents are not just an LLM calling tools in a loop. The harness — memory, skills, safety, context control, delegation, and extensibility — is what separates a demo from a production system.
4
5	For: Engineers building or extending coding-agent runtimes, custom agents, or advanced multi-agent workflows.
6	Not for: Prompt engineering, model selection, generic software architecture, or LLM API basics.
7
8	All principles are distilled from production runtime decisions. Claude Code is used as grounding evidence, not as the only possible implementation.
9
10	## Choose Your Problem
11
12	\| If you want to... \| Read \|
13	\|---\|---\|
14	\| Make the agent remember and improve over time \| [Memory](#1-memory) \|
15	\| Package reusable workflows and expertise \| [Skills](#2-skills) \|
16	\| Let the agent use tools powerfully but not dangerously \| [Tools and Safety](#3-tools-and-safety) \|
17	\| Give the agent the right context at the right cost \| [Context Engineering](#4-context-engineering) \|
18	\| Split work across multiple agents without losing control \| [Multi-agent Coordination](#5-multi-agent-coordination) \|
19	\| Extend behavior with hooks, background tasks, or startup logic \| [Lifecycle and Extensibility](#6-lifecycle-and-extensibility) \|
20
21	Before you start building: Read the [Gotchas](#gotchas) — these are the non-obvious failure modes that cost the most time.
22
23	---
24
25	## 1. Memory
26
27	User problem: "My agent forgets corrections and project rules between sessions."
28
29	Golden rule: Separate what the agent knows (instruction memory) from what the agent learns (auto-memory) from what the agent extracts (session memory). Each layer has different persistence, trust, and review needs.
30
31	When to use: Any agent that operates across multiple sessions or needs to accumulate project-specific knowledge over time.
32
33	How it works:
34
35	- Instruction memory is curated, hierarchical configuration injected into system context in priority order (org-wide → user → project → local; local wins). This is where project conventions, coding standards, and behavioral rules live. It is human-authored and stable.
36	- Auto-memory is agent-written persistent knowledge with a type taxonomy (user / feedback / project / reference) and a capped index. Saving is two-step: write a topic file, then update the index. The cap prevents unbounded growth — without cleanup, recent entries silently disappear.
37	- Session extraction runs as a background agent at session end. It directly writes to auto-memory — topic file then index — following the same two-step save invariant. A mutual-exclusion guard ensures that if the main agent already wrote memory during the turn, the extractor skips entirely. This is the autonomous learning loop.
38	- Review and promotion audits across all memory layers and proposes cross-layer moves (auto-memory → project conventions, personal instructions, or team memory). It never applies changes autonomously — proposals require explicit user approval.
39
40	Start here: Define your memory layers (instruction, auto, extraction). Implement the two-step save invariant (topic file, then index). Add background extraction only after the core write path is stable.
41
42	> In Claude Code: Use `/remember` to audit and promote auto-memory entries across layers.
43
44	Tradeoffs:
45
46	- More memory layers = richer recall but higher maintenance burden. Without periodic pruning, index caps cause silent data loss.
47	- Session extraction adds latency at session end but dramatically improves cross-session learning.
48
49	Go deeper: [references/memory-persistence-pattern.md](references/memory-persistence-pattern.md)
50
51	---
52
53	## 2. Skills
54
55	User problem: "I want my agent to reuse workflows and domain knowledge without re-explaining them every time."
56
57	Golden rule: Skills are lazy-loaded instruction sets, not eagerly injected prompts. Discovery must be cheap (metadata only); the full body loads only on activation.
58
59	When to use: Any agent that needs reusable, composable workflows activating on matching user intent.
60
61	How it works:
62
63	- Discovery is budget-constrained: the agent sees a compact listing of all available skills (name, description, and when-to-use hint concatenated per entry), each hard-capped at a fixed character limit, with the total capped at roughly 1% of the context window. Front-load your trigger language — tails get truncated.
64	- Loading is lazy: only metadata enters the always-on context. The full skill body loads only when the skill activates, k

People who liked this also viewed

6 picks

Type