Off-Plan Feature Log

    1

    get added halfway through a build, designs get simpler, and new guards get added after something breaks.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Off-Plan Feature Log

    Example session with this skill installed

    The CSV export and the retry-on-timeout guard are done and tested. Log them in the feature log.

    • Read your context and instructions
    • Compiled the off-plan feature log

    Summary

    I checked both features against PLAN.md. The CSV export is covered by PLAN §3.2, but it was built differently, so I logged only the difference (F7). The retry guard isn't in the plan at all, so it gets a full off-plan entry (F8).

    Added to FEATURES.md

    F7. CSV export (differs from PLAN §3.2)

    • Added: 2026-11-09 · Status: done · Off-plan: differs from PLAN §3.2

    What it is: PLAN §3.2 specified an async export job with an email link. We built a direct streamed download instead.

    • Why we built it: exports are under 5,000 rows in practice, so a job queue added moving parts for no benefit.

    How it works: 1. GET /reports/{id}/export calls stream_csv() in reports/export.py.
    2. Rows are written in chunks of 500, so memory stays flat.

    Decisions and trade-offs: no email link, and very large exports will be slow. We can return to the async job if exports grow.

    • Limits: no Excel format and no column picker.
    • Where: reports/export.py, api/routes/reports.py
    • Proof: tests/test_export.py (4 tests) [verified]
    • Not verified: exports above 50,000 rows

    Likely questions: "Why not the async job from the plan?" Real exports are small, so streaming is simpler. "What if exports grow?" The route can switch to the job without changing the client.

    F8. Retry-on-timeout guard

    • Added: 2026-11-09 · Status: done · Off-plan: yes
    • What it is: calls to the payment provider are retried up to 3 times with backoff when they time out.
    • Why we built it: staging showed intermittent provider timeouts that failed whole checkouts.

    How it works: 1. with_retry() in payments/client.py wraps each call.
    2. Only timeouts are retried, never declines.

    • Limits: no circuit breaker yet.
    • Where: payments/client.py
    • Proof: tests/test_retry.py [verified]
    • Not verified: behaviour under a full provider outage

    Recent additions (top of the file)

    • 2026-11-09 · F8 · Retry-on-timeout guard for payment calls (off-plan)
    • 2026-11-09 · F7 · CSV export streams directly instead of the planned async job

    Nothing was committed.

    (Example project. The plan filter and entry template are exactly what the skill uses.)

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

    About this skill

    The problem

    Plans drift. Features get added halfway through a build, designs get simpler, and new guards get added after something breaks. A few weeks later, nobody can say what was built beyond the plan, or why.

    What it does

    After a feature is finished and tested, Claude:

    1. Finds your plan file (PLAN.md, SPEC.md, PRD.md, ROADMAP.md or docs/plan*.md).
    2. Searches the plan for the feature and applies a plan filter:
      • Built as planned: nothing is logged. Claude tells you which plan section covers it.
      • Built differently or beyond the plan: only the difference is logged, with a reference to the plan section.
      • Not in the plan at all: a full entry is logged and marked Off-plan: yes.
    3. Reads the real code and tests, and writes or updates one entry in FEATURES.md. It never adds duplicates.

    Each entry covers what the feature is, why it was built, how it works step by step, the decisions and trade-offs, its limits, where the code lives, proof (tests or runs), what was not verified, and two likely questions with honest answers.

    Together, your plan and FEATURES.md describe the system as it really is.

    Built-in honesty rules

    • Every claim must come from code Claude read or a test it ran, never from memory of the chat.
    • Anything not verified goes in a Not verified line.
    • No invented numbers. Metrics are copied only from real runs.
    • No API keys, tokens, emails or private paths in the file.
    • Nothing is committed unless you ask.

    Good for

    • Freelancers and agencies: filter on Off-plan: yes to get the list of work beyond the agreed scope for a change request.
    • Teams: handovers and onboarding, with one file that explains everything the spec doesn't.
    • Solo builders: explaining your own project in a review, demo or interview.

    How to use

    Run /off-plan-feature-log after finishing a feature, or say "log this feature" or "what did we add outside the plan?". It works with any agent that supports SKILL.md and can read files and run grep.

    Limits

    • It needs a written plan to compare against. Without one, it asks which file to use.
    • It logs finished, tested work only. Refactors and docs-only changes are skipped on purpose.

    Skips refactors and docs-only changes. Requires an existing PLAN.md, SPEC.md, or similar. Only logs finished and tested work.

    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 any payment.

    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 forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 7 days ago

    • Free with an Agensi account

    Permissions

    Read Files
    Terminal / Shell
    Write Files

    Instructions only, no scripts. Read files: reads the plan file and the code and tests of the finished feature. Shell: runs grep to search the plan. Write files: creates or updates FEATURES.md in the repo root. No network access, nothing is committed unless the user asks.

    Listed7 days ago

    What's inside

    Frequently Asked Questions

    The skill identifies deviations by comparing your actual code and test results against your `PLAN.md` or `SPEC.md`, documenting any features or architectural trade-offs that weren't in the original scope.

    This skill is compatible with any AI agent capable of file system operations and command-line tool execution, and it works with any codebase that uses Markdown for project planning.

    The skill requires an existing plan file (like `PLAN.md` or `ROADMAP.md`) in your repository root or docs folder to act as a baseline, along with a `FEATURES.md` file for the output.

    Unlike manual prompts, this skill enforces 'verification tracking' by cross-referencing claims with actual code reads and test outputs to prevent documentation hallucinations.

    It specifically tracks new features and architectural trade-offs; it intentionally ignores code refactors, variable renames, or purely stylistic documentation changes.

    Your purchase includes the skill definition, a standardized logging template, and the logic required to perform diff identification and verification across your Markdown files.

    New in Agents & Orchestration