agentsclimarketplace

Build agent harnesses

Skill hiteshbandhu/skills-i-use/skills/ai-engineer-talks/build-agent-harnesses

Runs checklists and workflows for designing, hardening, and operating agent harnesses — guardrails, verify steps, tool loops, durable sessions, eval, quotas, coding/deep-research/voice harnesses. Use when the user builds agent runtimes, asks what a harness is, scopes agent vs workflow, or says "agent harness", "guardrails", "verify step", "FOMAT", "durable session".From its SKILL.md

Install
npx -y skills add hiteshbandhu/skills-i-use --skill build-agent-harnesses

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 3 commands, including `cp -r skills/build-agent-harnesses ~/.claude/skills/` and 2 more.

SKILL.md

3.4 KB, 788 tokens by cl100k_base, as published. Nobody here has run it

Build agent harnesses

Action playbook from twenty-two AI Engineer talks on agent & harness engineering. Do not summarize talks — pick a workflow and execute it.

Supporting files (read when needed):

  • workflows.md — workflows A–L (steps, deliverables, stop conditions)
  • source-index.md — src-NNN → talk learnings in ingest-into-skills

Optional deliverables: {SKILL_OUTPUT_DIR}/build-agent-harnesses/ — see skills-i-use OUTPUT.md.


Step 0 — Pick workflow

Use the decision tree below. Open the matching section in workflows.md.

What is the user trying to do?
├─ Define harness vs agent loop + guardrails/verify     → A
├─ Decide agent vs workflow for a use case              → B
├─ Maturity path: framework → state machine → scale     → C
├─ Enterprise: quotas, skills, multi-agent platform     → D
├─ Agent evaluation + anti-hype discipline            → E
├─ Long-horizon research / deep research harness        → F
├─ Coding agent / computer-use / IDE harness            → G
├─ Production chat UX: resume, steer, multi-device      → H
├─ Operate many parallel coding agents (FOMAT)          → I
├─ Regulated / domain-vertical knowledge agents         → J
├─ Delete scaffold; maximize LLM compute in backend     → K
└─ Post-training / finetune / RL for tool agents        → L

Stop summarizing once a workflow is identified — run its checklist.


Install

cp -r skills/build-agent-harnesses ~/.claude/skills/
cp -r skills/build-agent-harnesses ~/.cursor/skills/
cp -r skills/build-agent-harnesses ~/.codex/skills/

From skills-i-use or ingest-into-skills after sync.

Source corpus: ingest-into-skills playlists/agent-harness-engineering-ai-engineer/.


Cross-cutting rules

RuleSource
Harness = tools + context + guardrails + verify around the model[src-001 @ 4:36]
Do not prompt harder when verify shows failure — fix harness[src-001 @ 9:37]
Agents only when ambiguity/value/error profile justify cost[src-002 @ 2:57]
Every agent is a state machine; prune prompts as models improve[src-003 @ 4:37]
Agent eval needs environments + actions, not I/O strings only[src-006 @ 7:52]
Decouple clients via durable sessions for resume/steer[src-015 @ 5:25]

Disputed steps: read talk in source-index.md.


Output to user

  1. Name the workflow (A–L) and what you are producing
  2. Save artifacts under ./skill-outputs/build-agent-harnesses/ when the user wants files
  3. Do not auto-commit

Invocation examples

@build-agent-harnesses design a harness for our browser agent
agent vs workflow for this support bot?
we need quotas and verify steps for production agents

What ships with it: 3 files

14.2 KB alongside SKILL.md

Gives 0 of the 12 instructions most context ai engineering skills give in 788 tokens

Counted across 1,328 of the 2,349 authors here whose files we hold, read 2026-09-06

  • Dispatch a fresh subagent for each taskin 76 of 1328, across 59 files
  • Perform spec compliance review before code quality reviewin 44 of 1328, across 34 files
  • Dispatch a final code reviewer after all tasksin 38 of 1328, across 26 files
  • Answer subagent questions before allowing implementationin 36 of 1328, across 26 files
  • Use the least powerful model capable of the taskin 33 of 1328, across 26 files
  • Create a TodoWrite list for all tasksin 32 of 1328, across 22 files
  • Perform a task review after each implementationin 31 of 1328, across 24 files
  • Extract all tasks and context from the planin 29 of 1328, across 20 files
  • Provide full task text to subagentsin 28 of 1328, across 20 files
  • Use git worktrees for isolated workspacesin 25 of 1328, across 20 files
  • Specify the model explicitly when dispatching a subagentin 23 of 1328, across 18 files
  • Execute all tasks from the plan without stoppingin 21 of 1328, across 16 files

Said here and by no other author read

  • Pick a workflow from the decision tree
  • execute the workflow checklist
  • read supporting files when needed
  • fix the harness when verify steps fail
  • decouple clients using durable sessions
  • name the workflow and output produced

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.