agentleFS
kimgoetzkekimgoetzke/coding-agent-configs/skills/improve-a-plan/SKILL.md

improve-a-plan

Skill2 starsChanged 7 months agoMITView on GitHub

Review, criticise and/or test an existing plan with the goal of improving is. Use when user asks you to review or improve a plan.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    No output: use local-ai-folder-markdown, the default for this skill group

    line 16
  2. Very specific:

    Read ../planning/persistence/{implementation}.md and follow it whenever this skill lists or opens a plan, or reads, writes or updates a …

    line 19
  3. Distinctive:

    If you haven't already, read the planning documents now

    line 33
  4. Distinctive:

    If the user response with yes, give the user the following options to choose from:

    line 56
  5. Distinctive:

    If no, encourage the user to start a fresh session

    line 60
  6. Distinctive:

    If you are in read-only mode, prompt the user to disable it in order to make these changes

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Persistence

This skill belongs to the plans skill group. A persistence implementation decides where and how plans are stored. Before locating the plan, find the setting:

R="$(git rev-parse --show-toplevel 2>/dev/null)/.ai/config.properties"; { cat "$R" ~/.ai/config.properties | grep '^persistence\.plans[[:space:]]*=[[:space:]]*[^[:space:]]'; cat "$R" ~/.ai/config.properties | grep '^persistence[[:space:]]*=[[:space:]]*[^[:space:]]'; } 2>/dev/null | head -1 | tr -d ' \t\r'
  • No output: use local-ai-folder-markdown, the default for this skill group
  • persistence.plans={value} or persistence={value}: use {value}. If ../planning/persistence/{value}.md does not exist, tell the user the plans skill group has no {value} implementation and use local-ai-folder-markdown

Read ../planning/persistence/{implementation}.md and follow it whenever this skill lists or opens a plan, or reads, writes or updates a planning document.

Phase 1: Locate the plan

Resolve the target plan using the first matching strategy:

  1. Context inference — check your conversation context for a plan location from a prior planning-mode or planning skill invocation in this session. If found, use it directly and confirm with the user (e.g. "Improving plan 2026-03-24 auth refactor — correct?"). Proceed on confirmation; if the user says no, fall through to the next strategy.
  2. User argument — if the user provided an argument, resolve it as a number or plan name against the listed plans e.g. 2026-01-20 improve-mfa-flow. Do not read any plan documents until the plan has been resolved.
  3. Single plan — if only one plan exists, select it automatically.
  4. Multiple plans — list plans newest first in a numbered list and ask the user to choose. Do not read any plan documents until the plan has been resolved.
  5. No plans — tell the user no plans were found and suggest creating one with the planning-mode or planning skill.

Phase 2: Improve the plan

  • If you haven't already, read the planning documents now
  • Tell the user that they generally have the following options:
1. Adversarial review in a fresh session

Start a new session and ask me to critique the plan with fresh eyes (no bias from having written it).
Something like: "Read the plan in <location of plan> and the findings. Identify assumptions, gaps, and areas likely to break during implementation." A separate session has clean context and is more likely to spot issues than the one that wrote the plan.

2. Dry-run in plan mode

Use Shift+Tab to switch to plan mode, then ask me to walk through the plan as if implementing it — reading the actual
source files, checking types exist, verifying method signatures match what the plan assumes. This surfaces mismatches
between the plan and reality (e.g. a method signature that's slightly different, a missing dependency, a type that
doesn't exist yet).

3. Explicit assumption check

Ask me to list every assumption the plan makes (e.g. about an external API) and verify each one against an actual
source. The findings doc has some of this, but after a lot of back and forth, it may be worth a refresh.
  • With the above explanation, provide your own assessment of which option will likely be the highest value next step and ask the user if they would like to continue with the current session (yes/no)
  • If the user response with yes, give the user the following options to choose from:
    1. Complete an adversarial review
    2. Dry-run the plan
    3. List every assumption and verify it against the actual source(s)
  • If no, encourage the user to start a fresh session
  • Once you have completed either of the options that can improve a plan
    • You must update the findings document and update or create, if it does not exists, the questions document with your findings
      • If you are in read-only mode, prompt the user to disable it in order to make these changes
    • The changes you make to the questions document should be your findings/concerns/challenges phrased as clarification or decision questions to the user
    • You must encourage the user to review or discuss these findings
    • You must update all other plan documents accordingly, if necessary

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.