specification-coach

A question-led guide that turns a vague feature request, issue, or coding task into a clear implementation specification. The specification describes observable behaviour, limits, failure cases, and conditions for acceptance.

In plain words
What is it for?
Use it to clarify requirements, define what is included and excluded, set constraints, describe failure behaviour, and produce a small verifiable starting scope.
Why use it?
It prevents an agent or developer from filling in important product decisions differently from what you intended.

Skill for Claude CodeCodex

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 skills/far-200/think-before-code/specification-coach
Any agent
npx skills add Far-200/think-before-code --skill specification-coach
Clone the repo
git clone --depth 1 https://github.com/Far-200/think-before-code

Made for: Claude Code, Codex.

Per session 166 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 3,982 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 $0.00166 $0.03982
Opus 5 $0.00083 $0.01991
Sonnet 5 $0.00033 $0.00796
Haiku 4.5 $0.00017 $0.00398

Measured 2d ago against content hash 6feb71d9b687, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

specification-coach 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/specification-coach/SKILL.md · 478 lines

How it starts

The opening of the file, as written. The whole thing — 478 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Specification Coach

Why this exists

Agents are now asked to implement things. The bottleneck moved.

Hand an implementation agent "add CSV export to the orders page" and it will not stop. It infers the missing requirements — every field, every permission, a background job, a retry policy — and starts coding against a product nobody specified. The code arrives. Whether it is the right code is unanswerable, because nothing was defined well enough to check it against.

The failure is upstream of the code. A request that two competent developers could implement differently while both claiming success was never a specification; it was a wish. This skill exists to make the learner do the defining: resolve one ambiguity at a time, own each decision, and turn those decisions into observable acceptance criteria — so the specification is theirs, and so is the judgment about what a specification needs to contain.

The contrast this skill enforces:

Traditional assistant:
vague request → inferred requirements → generated specification/code

specification-coach:
vague request → one ambiguity → learner decision →
observable criterion → learner-authored specification → handoff

This is not a PRD generator, project manager, system designer, prompt engineer, or implementation agent. Its job is specification judgment.

Boundary with neighbouring skills

  • vs. code-review-coach — this skill operates before implementation, when the desired behaviour is not yet pinned down. code-review-coach begins once code, a diff, or a pull request exists and the learner wants to discover risks in it. After implementation, hand the completed work to code-review-coach. Never review hypothetical code during a specification session.
  • vs. test-case-coach — acceptance criteria define what a future feature must observably do. test-case-coach designs a compact executable suite for an approach or implementation that already exists. A few verification examples may live in a specification; a systematic suite is a handoff.
  • vs. debug-coach — undefined or disputed desired behaviour is a specification problem. A concrete expected-vs-actual failure on a specific input is a debugging problem. Once implementation produces an observed failure, hand off to debug-coach.
  • vs. problem-decoder — problem-decoder reads a DSA or LeetCode-style problem statement before solving. This skill resolves ambiguous real-world feature, product, workflow, or software change requests. Do not route ordinary LeetCode statements here.
  • vs. dsa-tutor — dsa-tutor builds an algorithmic solution. This skill defines observable software behaviour before implementation. Do not turn DSA solving into product-specification theatre.

Read the full file on GitHub · 478 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 · 478 lines · 166 tokens per session scan A 6feb71d9b687

Subscribe to this mod's changes

specification-coach is a skill published in the GitHub repository Far-200/think-before-code (8 stars, last pushed 18d ago), licensed MIT. It adds 166 tokens to every session and 3,982 once invoked, about $0.0008 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 2026-08-31.

Related

Other skills, from other repositories

interactive-leetcode-mcp

Use when the user wants to practice LeetCode problems, submit solutions, or set up LeetCode integration. Covers MCP server installation, learning-guided practice flow, solution submission, and authentication.

SPerekrestova/interactive-leetcode-mcp · 46 tokens

leetcode-helper

Expert LeetCode problem solver for optimized C++ coding interview solutions. Use when: the user pastes a LeetCode problem, asks for an optimized algorithm, wants C++ LeetCode code, needs a dry run, asks for line-by-line explanation, wants a hint before the solution, wants brute force to optimized progression, wants…

Zephyrex21/claude-leetcode-helper · 98 tokens

configure-profile

Create or update the global learner profile. Interviews the learner using the measurement checklist in references/PROFILE-TEMPLATE.md and writes the structured result to profile/PROFILE.md. Use when a new learner sets up Upstack for the first time, or after completing a course to update skills and Dreyfus levels.

ishands/upstack · 64 tokens

create-course

Scaffold a new course directory from the COURSE.md schema. Creates the directory structure, populates COURSE.md with YAML frontmatter and markdown template sections, and creates stub assignment directories. Use for curated courses (core/courses/) or personal courses (custom/courses/) in any domain.

ishands/upstack · 60 tokens

start-course

Initialise a learning session. Checks for an existing journal in progress/ /journal.md, creates one from the template if missing, loads the course definition, and calibrates to the learner's declared context. Use when starting a new course or resuming after a break.

ishands/upstack · 60 tokens

start-learning-path

Enrol in a curated learning path. Discovers available paths, presents project options in plain language, records the learner's choice, initialises the integration journal, and displays the full course roadmap. Run once at the start of the path — not at the start of each course. Distinguishes enrolment (path level…

ishands/upstack · 83 tokens