Forge Spec: Technical Design Before Anyone Codes

    1

    Turn approved requirements into a technical spec with exact interfaces, a file map and resolved decisions, so every implementer builds against the same contract.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Forge Spec: Technical Design Before Anyone Codes

    Example session with this skill installed

    Turn ticket.md into a proper PRD.

    ticket.md:

    Hey, can we get a little command-line habit tracker? Something like habit add "read",
    habit done read and habit streak read. It should remember stuff between runs and
    be pleasant to use. Maybe show a weekly view too. Should work offline obviously.

    • Read your context and instructions
    • Compiled the forge spec

    The ticket has a clear shape but no Non-Goals and several untestable phrases ("pleasant to use", "remember stuff", "weekly view"). Before writing, here are the gaps, all in one message:

    Non-Goals are missing. Proposed: local-only CLI; no accounts, sync, reminders, GUI or analytics beyond the requested views.
    2. Failures: how should duplicate or unknown habits behave? Proposed: exit non-zero with a clear error.
    3. Streak: what counts? Proposed: consecutive daily completions ending today; no completion today means 0.
    4. Weekly view: which week? Proposed: current Monday-Sunday, X marks a completed day.
    5.

    "Pleasant to use" can't be checked by a command. Proposed: a documented --help (testable) plus one [manual-check] for wording.

    (User: "all defaults OK")

    Wrote docs/forge/prd/001-habit-tracker-cli.md: 5 user stories, 11 Given/When/Then criteria, Non-Goals, and an Adjustments changelog.

    
    # PRD-001: habit-tracker-cli
    
    ## Goal
    Provide a small offline command-line habit tracker that lets a person add habits, record daily completions, review streaks, and see the current week. Habit data must persist locally between separate command invocations.
    
    ## Non-Goals
    - No user accounts, authentication, cloud sync, or multi-device sharing.
    - No reminders, notifications, or graphical user interface.
    - No habit scheduling, goals, or analytics beyond daily completions, streaks, and the current-week view.
    
    ## User Stories
    ### US-1: Add a habit
    As a person tracking habits, I want to add a named habit, so that I can begin recording it.
    - AC-1.1: Given no habit named "read" exists in an empty local data store, When I run `habit add "read"`, Then the command exits with status 0, prints a confirmation containing `read`, and a separately started `habit streak read` command exits with status 0 and reports a streak of 0.
    - AC-1.2: Given a habit named "read" already exists in the local data store, When I run `habit add "read"`, Then the command exits with a non-zero status and does not create a second `read` habit.
    
    ### US-2: Record a daily completion
    As a person tracking habits, I want to mark a habit done for today, so that my progress is retained.
    - AC-2.1: Given a habit named "read" exists and the computer date is 2026-10-22, When I run `habit done read`, Then the command exits with status 0, prints a confirmation containing `read`, and a separately started `habit streak read` command reports a streak of 1.
    - AC-2.2: Given no habit named "read" exists, When I run `habit done read`, Then the command exits with a non-zero status and prints an error that contains `read`.
    
    ### US-3: View a habit streak
    As a person tracking habits, I want to see my current streak for a habit, so that I can understand my consistency.
    - AC-3.1: Given the local data store records `read` as done on 2026-10-20, 2026-10-21, and 2026-10-22, and the computer date is 2026-10-22, When I run `habit streak read`, Then the command exits with status 0 and prints `3` as the current streak.
    - AC-3.2: Given the local data store records `read` as done on 2026-10-20 and 2026-10-22, and the computer date is 2026-10-22, When I run `habit streak read`, Then the command exits with status 0 and prints `1` as the current streak.
    - AC-3.3: Given the local data store records `read` as done on 2026-10-21 but not on 2026-10-22, and the computer date is 2026-10-22, When I run `habit streak read`, Then the command exits with status 0 and reports a streak of 0.
    
    ### US-4: See the current week
    As a person tracking habits, I want a weekly view, so that I can see which days I completed each habit.
    - AC-4.1: Given habits `read` and `walk` exist, `read` is recorded as done on 2026-10-20 and `walk` is recorded as done on 2026-10-22, and the computer date is 2026-10-22, When I run `habit week`, Then the command exits with status 0 and prints rows for both habit names, with `X` under the 2026-10-20 column for `read` and under the 2026-10-22 column for `walk`.
    - AC-4.2: Given the computer date is 2026-10-22, When I run `habit week`, Then the output identifies every date from Monday 2026-10-19 through Sunday 2026-10-25.
    
    ### US-5: Use the CLI comfortably
    As a person using the tracker, I want clear command guidance and feedback, so that I can use it without consulting source code.
    - AC-5.1: Given the CLI is installed, When I run `habit --help`, Then the command exits with status 0 and its output documents the `add`, `done`, `streak`, and `week` commands.
    - AC-5.2: Given successful and invalid invocations of the documented commands, When a person reviews their terminal output, Then the wording is clear and concise. `[manual-check]`
    
    ## Constraints
    - The tracker must be usable entirely from a command-line interface.
    - It must store its data locally and retain it across separate command invocations.
    - It must work without network access.
    
    ## Definition of Done
    - [ ] All AC green
    
    ## Adjustments
    - **ID remap:** `habit add` -> US-1; `habit done` -> US-2; `habit streak` -> US-3; weekly view -> US-4; "pleasant to use" -> US-5.
    - **Reformulated:** "remember stuff between runs" -> AC-2.1; "Should work offline" -> Constraints; "weekly view" -> AC-4.1, AC-4.2; "pleasant to use" -> AC-5.1, AC-5.2.
    - **Added:** Non-Goals; duplicate and unknown-habit errors (AC-1.2, AC-2.2); a current-streak definition (AC-3.1 to AC-3.3); Monday-Sunday scope and `X` marker for the weekly view.
    - **Flagged [manual-check]:** AC-5.2.
    - **Dropped:** none.
    

    Next step?
    a) Technical design first (forge-spec, if installed) —

    Recommended: the CLI interface and the on-disk data format need to be agreed before coding
    b) Straight to a phase plan (forge-plan, if installed)
    c) Stop here — the PRD is the deliverable

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Turn an approved PRD into the HOW that sits between requirements and code. Decisions get made once, so whoever implements (a person, an agent, or several agents in parallel) never re-derives the architecture in incompatible ways.

    What you get

    • Architecture in 3-10 lines: the smallest design that satisfies every acceptance criterion. No speculative generality.
    • Components: one responsibility each, tied to the user stories they serve. A component that serves no story gets deleted.
    • Exact interfaces: function signatures, request/response schemas, routes, DDL, CLI contracts. Not prose.
    • File map: every file to create or modify, with its role and component.
    • Decisions: every trade-off resolved, one line each with the reason.
    • Risks: each mapped to the components and criteria it threatens, with a mitigation.

    How it works

    1. Reads the approved PRD, your repo's structure and conventions, and past decisions in docs/context/ if present. The spec extends what exists; it never invents a parallel architecture.
    2. Tells you when a spec isn't worth it (small task, obvious structure) and lets you skip it.
    3. Fills the template, then checks coverage: every user story is served by a component, every needed interface is specified.
    4. Anything it can't decide goes out in one batched message with a recommended default for each question. No TBD survives.
    5. Writes docs/forge/specs/spec-NNN.md and stops at your approval.

    Good fit when

    • Several components, people or agents must agree on interfaces.
    • You are introducing a data model, an API or a file format.
    • You want a risk list that tells a planner which parts are not "simple".

    Requirements

    Any SKILL.md-compatible agent. No scripts, no network, no dependencies. Works best with a PRD that has stable IDs (US-n, AC-n.m), such as one from forge-prd.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.

    Requires a PRD with stable IDs. Does not generate implementation code. Scoped to design documentation only.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 20 days ago

    Permissions

    Read Files
    Write Files

    Reads the requirements doc you point it to (and docs/context/lessons.md or memory.md if they exist). Writes one Markdown file under docs/forge/prd/. No shell commands, no network, no environment variables.

    Listed20 days ago

    What's inside

    Frequently Asked Questions

    It prevents 'design drift' by converting vague requirements into rigid technical contracts, ensuring that every function signature, API route, and database schema is defined before a single line of code is written.

    Yes, while it integrates natively with other Forge tools, it can ingest any markdown-based PRD or requirements document provided it has clear user stories or feature IDs to map against.

    The skill produces standard Markdown documentation stored in your local repository, making it compatible with any IDE, version control system, or AI agent that can read Markdown files.

    Forge Spec identifies technical trade-offs and 'TBD' areas during the drafting phase, presenting them as a batched list for you to resolve before it finalizes the technical design.

    No, this skill focuses strictly on the 'How' of the design; it generates the blueprint and file maps, but does not execute the actual coding or refactoring of your source files.

    Upon purchase, you receive the full agent skill package, including the structured spec templates and the logic required to enforce interface coverage and component mapping.

    New in Prompt & Skill Engineering