More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Project Docs: Interview to PRD, SRS & Architecture

    1

    Most projects start coding before anyone writes down what to build, and AI agents fill the gaps by guessing.

    Free

    Security scanned
    Project Docs: Interview to PRD, SRS & Architecture

    Project Docs: Interview to PRD, SRS & Architecture

    Example session with this skill installed

    /project-docs A mobile app for tracking local surf conditions using community reports and weather APIs. Targeted at weekend hobbyists. Must work offline.

    • Read your context and instructions
    • Compiled the project docs
    • Generated the document

    A real run on the request above, with every question answered default. About 40 minutes start to finish. Below is the skill's final review (Confirm Gate 4), then the feature table it wrote to docs/PRD.md.

    Review results

    Consistency and traceability: every Must feature has user stories and every story has acceptance criteria. Every functional requirement maps to a story and a test, and every non-functional requirement has a number and a way to verify it. No orphans, no undefined IDs, no broken links.

    Writing rules: no banned vague words, no em or en dashes, no curly quotes, no placeholders. Every unknown is in the Open Questions table.

    Cold-read test: run twice with an agent that had no context. Both times it answered all six questions correctly. It also found 29 gaps in total (16, then 13), all fixed. The most important fixes:

    Task order: the code that fetches the live forecast data now comes before the M1 accuracy test that needs it (T-015 moved into M1).

    Agreement: a match now means the surfer's own Good/Fair/Poor rating for that hour equals the app's call. The M1 check only counts observations made after the last tuning change.

    • Break ids: CI now fails if an id is deleted from the catalog, so ids can't be silently reused.

    Versions pinned: Vitest 4.1.x (the Cloudflare test pool doesn't support Vitest 5), Maestro 2.12.0, gitleaks-action v3.0.0.

    Store accounts: a task now creates them, and a new task (T-047) runs Google Play's closed test, which new personal developer accounts have to complete before going live.

    Feasibility: the MVP fits one developer with no deadline. If time gets tight, cut F-08 (usage count) first; the store dashboards give rough numbers instead.

    Docs created: 21 files in docs/

    README, BRIEF, PRD, SRS, UX, ARCHITECTURE, TECH_SPEC, SECURITY, TEST_PLAN, PLAN, DECISIONS, LAUNCH, 8 feature specs in features/, and api/openapi.yaml.

    ItemCount
    Features15 (8 in the MVP, each with a spec)
    User stories18
    Acceptance criteria107
    Functional requirements51
    Non-functional requirements24
    ADRs23
    Tasks47, in 7 milestones
    Risks11
    Open questions12: 4 resolved, 3 using the default, 5 open

    Open questions that block the MVP

    #QuestionBlocks
    OQ-9Which phones do you have for testing? Default: your iPhone plus a used Pixel 7a-class Android.M0: the 3-second startup measurement
    OQ-4Do you already have Apple Developer and Google Play accounts?M0: putting test builds on an iPhone; launch
    OQ-11Who judges the app's calls in M1, and who are the 5 surfers for the M6 sessions?Leaving M1 and M6
    OQ-10Support email addressStore submission
    OQ-8Final app nameStore listing

    Feature table (docs/PRD.md)

    F-IDNamePurposePersonaPriorityRelease
    F-01Break catalogCurated data for 30 to 50 launch-region breaksAll, ownerMustMVP
    F-02Find and save breaksBrowse, search, sort by distance on the device, save up to 10AllMustMVP
    F-03Forecast fetch and offline cacheBuild the bundle every 3 h; cache it on the device; show its ageAllMustMVP
    F-04Call enginePer-hour Good/Fair/Poor call and good daylight windowsAllMustMVP
    F-05HomeSaved breaks with call now and next good windowP1, P3MustMVP
    F-06Break detail48 h strip, now card, tide curve, windowsP2, P3MustMVP
    F-07About and legalDisclaimer, data credits, privacy policyAllMustMVP
    F-08Usage countAnonymous weekly install count for G-01 and G-02OwnerShouldMVP
    F-09Units settingSwitch ft/m and mph/kn/km/hAllCouldv1.x
    F-10"Was this call right?"One-tap feedback to tune F-04P1Couldv1.x
    F-11Window alertsPush when a saved break has a good windowP1, P3Shouldv1.x
    F-12Home screen widgetSaved break's call without opening the appP1Couldv1.x
    F-13Custom breakPin plus facing direction, if the catalog is thinP1Couldv1.x
    F-14Buoy readingsReal offshore readings next to the forecastP1CouldLater
    F-15More regionsBeyond Southern CaliforniaAllCouldLater

    Next steps

    • Reply approve and every doc is marked Approved.
    • Run /project-bootstrap. It reads these docs, skips the questions they already answer, and sets up CLAUDE.md, skills, settings and infrastructure to match. Then it starts on the first task in docs/PLAN.md.
    • Nothing is committed. The skill suggests a commit message and asks before git init.

    project-docs-interview-to-prd-srs-archit.pdf

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

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

    About this skill

    Most projects start coding before anyone writes down what to build, and AI agents fill the gaps by guessing. project-docs runs a structured interview first, then writes the documents a new teammate and an AI coding agent both need: what you're building, exactly what it must do, how it's built, and the order to build it in. Works for web and mobile, new projects and existing codebases.

    What it does

    • Gated interview in three rounds (idea, scope, tech). Every question shows a proposed default; answer default to accept it.
    • Full docs set in docs/: BRIEF, PRD, SRS, UX, ARCHITECTURE, TECH_SPEC, SECURITY, TEST_PLAN, PLAN, DECISIONS (ADRs), plus one spec per feature. LAUNCH and an OpenAPI file when they apply.
    • Testable requirements: vague words like "fast" or "scalable" are banned unless followed by a number or a condition.
    • Stable IDs and traceability: F-01, US-001, FR-001, NFR-001, ADR-001, T-001, never renumbered. Every story maps to requirements and tests.
    • Mermaid diagrams for navigation, architecture, sequences and the data model, so they render on GitHub.
    • Build plan with milestones and tasks sized S/M/L, each small enough for one working session.

    How it works

    1. Detect: reads existing code and docs to pick new or existing project mode. Existing projects are documented as built, with each requirement marked Built, Partial or Not built.
    2. Interview: three rounds, each ending at a confirmation gate. You approve the brief, the feature list with the MVP line, and the stack before it moves on.
    3. Generate: writes the docs set, sized to the project. An MVP's SRS can be three pages.
    4. Review: checks for contradictions, orphan requirements and vague wording, then runs a cold-read test: a fresh agent with no context reads only the docs and must answer what you're building, what's in the MVP and what the first task is. Gaps are fixed before the final approval gate.

    Why this beats prompting it yourself

    A single prompt gives you one long, generic document. This skill keeps each fact in one place, linked by ID, so docs don't drift apart. Unknowns go into an Open Questions table instead of being guessed. It's re-runnable: it updates existing docs in place, keeps every ID, and shows the diff before overwriting.

    Use cases

    • Turning an idea into a scoped MVP that tests the riskiest assumption first.
    • Documenting an existing codebase before a refactor or a new feature.
    • A handoff package for a developer, contractor or AI coding agent.
    • The input for project-bootstrap, which turns these docs into Claude Code context (CLAUDE.md, rules, skills).

    Known limitations

    • Writes documentation only. It does not write application code.
    • Diagrams are Mermaid only.
    • Stops at four confirmation gates, so you need to be there to approve each round.
    • Built and tested with Claude Code. The cold-read review needs an agent that can run subagents.
    • Never commits, and asks before git init.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Business & Operations.

    Requires manual input at four specific confirmation gates. Diagrams are limited to Mermaid format. No calendar-based scheduling is provided.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    • Free with an Agensi account

    Software
    git initGit

    Permissions

    Read Files
    Write Files

    Writes only inside docs/, after you approve each gate. In an existing project it reads the codebase (manifests, routes, schema, CI config) to document what's built. Never commits; asks before git init. Records env var names only, never values.

    Listed19 days ago

    What's inside

    Frequently Asked Questions

    The skill provides a complete `docs/` directory including PRDs, Software Requirement Specifications (SRS), technical architecture with Mermaid diagrams, and a prioritized MVP roadmap. Every requirement is assigned a stable ID to ensure traceability across your project's lifecycle.

    Yes, the skill features a 'Detect' mode that analyzes existing codebases or legacy notes to create documentation that aligns with your current infrastructure while identifying gaps for future refactors.

    The documentation is output in GitHub-flavored Markdown and Mermaid.js, making it natively compatible with AI-powered editors like Cursor, Claude Code, and Windsurf to serve as a 'ground truth' for automated coding.

    The 'gated interview' consists of four mandatory approval stages—Idea, Scope, Technical Direction, and Consistency Review—preventing the AI from hallucinating requirements or proceeding without your explicit validation.

    Unlike generic prompts that often produce vague or repetitive text, this skill enforces strict ID persistence and technical justifications, ensuring every requirement is measurable, testable, and free of contradictions.

    New in Business & Operations

    Free