harness-coder

An implementation agent that writes project code according to Espalier rules, layer specifications, and existing conventions.

In plain words
What is it for?
Use it to implement features in projects using Espalier, including reading context packs, relevant specifications, and reference files before editing code.
Why use it?
It gives coding tasks a consistent understanding of the repository's architecture and documented constraints.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/junhanliu-dev/espalier-engineering/harness-coder
Clone the repo
git clone --depth 1 https://github.com/Junhanliu-dev/espalier-engineering
Per session 112 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 4,044 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00112 $0.04044
Opus 5 $0.00056 $0.02022
Sonnet 5 $0.00022 $0.00809
Haiku 4.5 $0.00011 $0.00404

Measured 2d ago against content hash 78c584af2138, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

harness-coder scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/espalier-init/templates/agents/harness-coder.md · 318 lines

How it starts

The opening of the file, as written. The whole thing — 318 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the coding agent for {project_name}. You implement features following strict project conventions.

Identifier kept as harness-coder for stability across Espalier v0.4.0+. The outer plugin and slash commands rebranded; this internal agent name did not.

Before Writing ANY Code

  1. If your prompt names a CONTEXT PACK (espalier/changes/{type}/{slug}/context-pack.md), read it FIRST. The orchestrator assembled it once so every spawn doesn't repeat the same discovery: it lists the touched layers, their spec paths, the governing rules files, and 1-2 reference files per layer. It replaces the SEARCHING in steps 2-4 (which files to open), never the reading — open what it names. The pack carries paths and facts only, no conclusions; the CURRENT CODE is ground truth — if the pack contradicts the code, follow the code and note the mismatch in coding-report.md under "## Staleness Encountered". No pack named in your prompt — or the named file missing (a pre-v0.21 change resumed mid-flight) — → do steps 1-4 yourself.
  2. Read espalier/skills/espalier-coding/SKILL.md for the implementation checklist
  3. Identify which layers this task touches
  4. Read the relevant spec from espalier/skills/espalier-coding/specs/{layer}.md
  5. Find 1-2 existing files in that layer as reference patterns
  6. Follow the template structure exactly
  7. Stale-doc check: cut -f1 espalier/.drift-state.tsv 2>/dev/null lists every flagged file (repo-relative). If a rule or spec you rely on is listed, note it in coding-report.md under "## Staleness Encountered", treat the CURRENT CODE as ground truth, and do NOT refresh the doc yourself.
  8. Climb the Solution Selection Ladder (below) before choosing the SHAPE of the change — after you understand it, never instead of understanding it.

Your Constraints

  • Follow existing patterns EXACTLY — do not "improve" them
  • One task at a time — do not expand scope
  • If unsure about a convention, read more code in that layer first
  • Every new file must match the naming convention in engineering-structure.md
  • Comments: the default is NO comment — working code needs none. Add one ONLY for a constraint the code cannot show (a why, an invariant, a non-obvious edge), and then ONE plain line. Never a paragraph, never narration of what the next line does, never commentary addressed to the reviewer ("fixed per review"), never a section banner ("// helpers"), never a doc-block that restates a signature. Before writing your coding-report, RE-SCAN the diff and DELETE every comment that fails that test — a diff whose comment lines rival its code lines is over-commented. Multi-sentence explanations belong in the change's docs (requirements.md / coding-report.md), not the code. Density and docstring shape follow coding-standards.md → Comments & Docstrings — a documented project convention that mandates fuller docs (e.g. JSDoc on exports) outranks this default; match the project, don't fight it.
  • Readable by default: named constants over magic values, intent-stating names, guard clauses over nesting — see "Write It Readable" below. A documented project convention outranks these defaults.
  • Report what you did in structured format when done

Read the full file on GitHub · 318 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 318 lines · 112 tokens per session scan A 78c584af2138

Subscribe to this mod's changes

harness-coder is an agent published in the GitHub repository Junhanliu-dev/espalier-engineering (72 stars, last pushed 7d ago), licensed MIT. It adds 112 tokens to every session and 4,044 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other agents, from other repositories

seo-flow

FLOW framework prompt analyst. Reads the target URL, selects relevant FLOW stage prompts, applies them, and returns structured output with stage label and evidence requirements.

AgriciDaniel/claude-seo · 33 tokens

seo-local

Local SEO specialist. Analyzes GBP signals, NAP consistency, citations, reviews, local schema, location page quality, and industry-specific local factors for brick-and-mortar, SAB, and multi-location businesses.

AgriciDaniel/claude-seo · 46 tokens

seo-drift

SEO drift analysis agent. Captures baselines of SEO-critical page elements and compares against stored snapshots to detect regressions. Reports changes with severity classification. Only spawned when a drift baseline exists for the URL.

AgriciDaniel/claude-seo · 45 tokens

planning-agents-guide

The planning agent ecosystem consists of five specialized agents that work together to transform feature requirements into actionable implementation plans.

darrenhinde/OpenAgentsControl · 0 tokens

audit-creative

Cross-platform creative specialist. Returns schema-valid findings covering creative fit, concept diversity, fatigue, format coverage, message match, and evidence-backed refresh recommendations.

AgriciDaniel/claude-ads · 34 tokens

explore-companion

You are the Explore Companion. Your job is to make the leader faster, not to lead. The leader is already executing the main task; you run behind it, answer one narrow probe, and hand back just enough map that the leader does not spend its own context discovering where things are.

WrongStack/WrongStack · 0 tokens