More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Plan Notes: Revise a Plan From Inline Notes in One Pass

    1

    Re-prompting in chat to fix one paragraph of a plan loses track of the paragraph. You leave notes on the plan's lines, and Claude answers all of them in one revision, logs what changed, and records the decisions.

    Free

    Security scanned
    Plan Notes: Revise a Plan From Inline Notes in One Pass

    Plan Notes: Revise a Plan From Inline Notes in One Pass

    Example session with this skill installed

    I left notes on the plan in plans/billing-migration.md. Revise it from my notes.

    • Read your context and instructions
    • Compiled the plan notes

    Claude runs plan-notes list and finds 4 notes. It changes the Rollout section to add a dry run, rewrites the Data section to name the source table, adds a rollback step, and declines one note with "no change: the vendor contract fixes the cutover date". It deletes the 4 @@ lines, appends a revision entry with one line per note, runs plan-decisions, and reports one REVERSAL: the new plan moves invoices before customers, which reverses an earlier decision. It points you to the viewer to read the diff.

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

    About this skill

    About this skill

    Reviewing an agent-written plan in chat means describing which part you mean, then checking that the fix did not break another section. A note placed on the block it is about removes the ambiguity, and answering all notes in one pass keeps the plan consistent.

    You write notes as lines starting with @@ under the block they are about, in any editor, or click + on a line in the local browser viewer. /plan-notes lists every note, answers each in the block it sits on, deletes the answered notes, and appends a revision entry with one line per note: what changed, or "no change" with a reason when Claude disagrees. It never answers a note in chat only. Then plan-decisions pulls the plan's decisions into a decision-ledger.md and flags any decision that reverses an active one. An outline pass labels each section's purpose and role and reports sections that are out of order. plan-gap cold-reads a plan and lists what a fresh session would still have to ask.

    The proof

    The README reports one measurement, from 2026-08-09 on one real plan: a session started cold from the plan had 13 blocking questions, and only 2 were missing rationale. The other 11 were identifiers: commit SHA, repo, branch, run commands, type shapes, and current state. plan-gap exists to catch those before you clear a session. The repo also has a before-and-after example (examples/csv-export.md and its revision) where a note asking whether an export only covers page 1 turns into a changed Approach section and a revision line.

    What you get

    SKILL.md with the revision procedure, outline pass, and revision entry format; bin/plan-notes.py (list, diff, and view), bin/plan-decisions.py (decision ledger), bin/plan-gap.sh (gap check), and bin/plan-notes-view.html (the split-diff viewer with a Structure tab). The scripts use only the Python 3 standard library.

    How to use

    1. Copy the bin files to ~/bin/ as plan-notes, plan-decisions, and plan-gap (no extensions), keep plan-notes-view.html beside plan-notes, make them executable, and add the skill to ~/.claude/skills/plan-notes/.
    2. Open a plan with plan-notes view plan.md and add notes with + or by typing @@ lines.
    3. Run /plan-notes in Claude Code, then read the diff in the viewer and click Mark read.

    Source and license

    Free to use and MIT licensed. The full source, install steps, and issue tracker are on GitHub at https://github.com/robertmnyborg/plan-notes.

    SKILL.md calls the scripts at ~/bin/plan-notes, ~/bin/plan-decisions, and ~/bin/plan-gap; this package ships them with .py and .sh extensions because the marketplace requires one, so copy them without the extension. The optional auto-save hook (hooks/save-plan.mjs, needs Node 18+) and the examples are in the GitHub repo, not this package. Decision extraction is a model estimate and costs one claude CLI call per run. plan-decisions uses fcntl file locking and does not run on Windows. The viewer has no authentication.

    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 8 days ago

    • Free with an Agensi account

    Requires

    Claude Code
    Python 3.9+
    bash
    The claude CLI on PATH (plan-decisions and plan-gap call `claude -p`)
    git (for `plan-notes diff`)
    A web browser for `plan-notes view`

    Permissions

    Allowed Hosts

    127.0.0.1 (local viewer only)

    File Scopes

    the
    plan
    file
    you
    review
    projects
    /
    project
    /decision-ledger.md
    (or
    $PLAN_NOTES_LEDGER)
    .
    .lastread
    snapshot
    beside

    The viewer is a local HTTP server on a random port with no auth. plan-decisions and plan-gap send the plan text to Claude through the claude CLI.

    Listed8 days ago

    What's inside

    Frequently Asked Questions

    You can paste feedback into chat, but the agent has to guess which block you meant and may change others. Notes on the block remove the guess, and the revision log shows one line per note so you can check every answer.

    Partly. The `@@` convention and `plan-notes list`, `diff`, and `view` work with any agent or editor. `plan-decisions` and `plan-gap` call the Claude Code CLI (`claude -p`), and the skill and optional auto-save hook are written for Claude Code.

    It binds to 127.0.0.1 on a random port with no authentication and exits 10 minutes after the tab closes. Do not expose it to a network.

    Extraction is a model estimate, so read the Unread reversals section. IDs, status changes, and file writes are done by code.

    On GitHub at https://github.com/robertmnyborg/plan-notes, under the MIT license. Every file in this package is there, plus the README and install steps.

    New in Productivity