eng-harness-in-a-box
The engineering-harness loop with zero dependencies — an engineering harness in a box.
Skill by AI-Substrate
Chalk Labs
Need help deploying eng-harness-in-a-box?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/ai-substrate-eng-harness-in-a-box
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/eng-harness-in-a-box
What's inside
- Skill name
eng-harness-in-a-box- Description
- The engineering-harness loop with zero dependencies — an engineering harness in a box. Use in any repo, especially one with no harness yet: find or establish a focal operating surface, name the deterministic proof before working, capture friction as you go, and offer one encodable improvement at the end. Advisory always — the user is never blocked, the agent is never silent. When a repo has an installed harness CLI and wants the full loop, use eng-harness-flow instead.
- License
- MIT
- Instructions
- 59 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# eng-harness-in-a-box
- We're going to do work. Assume friction will happen.
- Do the task — and leave the environment better than you found it.
- Find the one focal operating surface, or establish a tiny one. Use it, don't guess around it.
- Before work: how will done be proved, deterministically?
- Find or establish one neat capture home for observations.
- Capture friction as it bites, one line each, keep moving.
- At the end: magic wand? What did you infer that should have been proved?
- Human chooses. Encode the fix, not the memory.
- Repeat.
Guiding ethos: Don't apologise - fix.
\[found-or-established focal surface\] + \[deterministic proof\] + \[friction capture\] + \[human-chosen encoding\] = a harness, in a box.
---
## The steps, expanded
Two goals, every session: complete the requested task, and leave the
engineering environment better than you found it. The second never displaces
the first.
Every step follows one pattern: **find it, or establish it.** The repo may
have none of this yet — that is normal, not a blocker.
1. **Focal surface — find or establish.** Look for the one place this repo is
operated: a harness CLI, justfile, Makefile, scripts dir, package scripts.
Found: use it — don't guess around it. Missing: offer to start a tiny one
that wraps what already exists (build / test / run / lint). Wrap, don't
reinvent.
2. **Proof — find or establish.** Before working, ask: how will this be proved
done, deterministically? Name the sensors (build, tests, lint, typecheck,
boot, smoke). A missing sensor is worth a sentence out loud — and sometimes
ten minutes to add — never a silent workaround.
3. **Capture home — find or establish.** Discover where this repo already
keeps harness data (`.harness/`, a retro ledger, a notes file) and use it.
Missing: choose one neat home — `.harness/observations.md` for what humans
read (frictions, retros, magic wands), JSONL beside it only when entries
must be machine-read. One home, dated entries, never scattered notes.
4. **Work through the surface, capturing friction as it bites.** A misleading
error, an undocumented step, a retry, anything you had to infer that a
check could have proved: one line each in the capture home, then keep
moving. Don't apologise for the environment — it is a legitimate target
for work.
5. **Retro — two questions at the end, never skipped silently.** Record both
answers in the capture home:
- Magic wand: what one command, check, fixture, default, or error message
would make the next run easier, safer, or better proven?
- What did you have to infer that the harness should have proved?
6. **Encode — the human chooses.** Offer the best answer. If accepted, encode
the fix, not the memory: a command that does the thing, or a check that
proves it, beats a paragraph that explains it.
Discriminate: capture what the next person or agent would also hit; let
one-offs go. Never gate, score, or block — the paved path wins by being
easier, and a declined offer is a normal outcome, not a failure.From AI-Substrate/harness-engineering, MIT.
More from AI-Substrate/harness-engineering
- 0.529score
validate-v2
AI-Substrate
Adaptive, high-signal post-action validation for whatever was just produced — unified plan, tasks, code change, workshop, legacy spec, or any artifact.
SkillBFreshness: BClaude Code +9 - 0.525score
eng-harness-flow
AI-Substrate
Front door for the engineering-harness loop.
SkillBFreshness: BClaude Code +9 - 0.525score
builder
AI-Substrate
The single front door to the SDD pipeline (docs/plans/): research → product plan → implementation guide → tasks → implement → review → post-flight → ship.
SkillBFreshness: BClaude Code +9 - 0.524score
grill-agent-done
AI-Substrate
Interrogate and defend the definition of done and the evidence that will prove it — one claim at a time, until each claim is lined up against the right grade…
SkillBFreshness: BClaude Code +9 - 0.524score
telemetry-insights-narrate
AI-Substrate
Route the LLM edge of the cohort-telemetry insights layer: turn a generated insights.json into short prose for the report's reserved narrator slot (narrator)…
SkillBFreshness: BClaude Code +9 - 0.524score
flow-eval-run
AI-Substrate
Start a peer-driven flow-conformance eval run with zero archaeology: pick or scaffold a scenario, spawn a BLIND subject peer to do the task, then score the…
SkillBFreshness: BClaude Code +9 - 0.521score
plan-0-v2-constitution
AI-Substrate
Establish or refresh the project constitution and align the supporting norms documents before any planning phases begin.
SkillBFreshness: BClaude Code +9 - 0.521score
plan-v2-extract-domain
AI-Substrate
Collaboratively identify and formalize an existing codebase concept as a named domain.
SkillBFreshness: BClaude Code +9 - 0.519score
the-flow
AI-Substrate
Backwards-compatible redirect for the SDD pipeline.
SkillBFreshness: BClaude Code +9 - 0.515score
eng-harness-0-harnessability-assessment
AI-Substrate
Assess a Git repository's harnessability — how easily a human or agent can enter, operate, modify, observe, prove, and improve it through an engineering…
SkillBFreshness: BClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 42
- Last commit (this folder)
- Jul 31, 2026 (3 months ago)
- Last release
- Aug 30, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from AI-Substrate
- 0.529score
validate-v2
AI-Substrate
Adaptive, high-signal post-action validation for whatever was just produced — unified plan, tasks, code change, workshop, legacy spec, or any artifact.
SkillBFreshness: BClaude Code +9 - 0.525score
eng-harness-flow
AI-Substrate
Front door for the engineering-harness loop.
SkillBFreshness: BClaude Code +9 - 0.525score
builder
AI-Substrate
The single front door to the SDD pipeline (docs/plans/): research → product plan → implementation guide → tasks → implement → review → post-flight → ship.
SkillBFreshness: BClaude Code +9 - 0.524score
grill-agent-done
AI-Substrate
Interrogate and defend the definition of done and the evidence that will prove it — one claim at a time, until each claim is lined up against the right grade…
SkillBFreshness: BClaude Code +9 - 0.524score
telemetry-insights-narrate
AI-Substrate
Route the LLM edge of the cohort-telemetry insights layer: turn a generated insights.json into short prose for the report's reserved narrator slot (narrator)…
SkillBFreshness: BClaude Code +9 - 0.524score
flow-eval-run
AI-Substrate
Start a peer-driven flow-conformance eval run with zero archaeology: pick or scaffold a scenario, spawn a BLIND subject peer to do the task, then score the…
SkillBFreshness: BClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get eng-harness-in-a-box running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
eng-harness-in-a-box skill. Agent Mart. https://agentmart.dev/skills/ai-substrate-eng-harness-in-a-box (last updated 2026-10-24)