agentleFS
jelbirtjelbirt/LLM-Skill-Build/next-steps/SKILL.md

next-steps

Skill1 starsChanged 3 months agoApache-2.0View on GitHub

Figure out the logical next steps for the current project by delegating a token-efficient review to a subagent that reads the repo state, git history, and the most recent plan/roadmap docs, then returns a prioritized short list. Use whenever the user asks "what's next", "what should I work on", "next steps", "where did we leave off", "what's left", "help me plan the next session", or seems unsure how to move the project forward — even if they don't say "next-steps" explicitly.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the subagent finds no planning docs, its recommendations come from git history and code state alone — say so in the relay so the user …

    line 62
  2. Distinctive:

    If the user's focus argument conflicts with what the roadmap prioritizes, present both rather than silently following one.

    line 64

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “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

Next steps

Determine the most productive next steps for this project and report them as a short prioritized list — without burning main-session context on the review.

Instructions

  1. Distill the current conversation into at most 5 bullets: the project's goal, decisions made this session, work completed, and known blockers. If the session is fresh with no meaningful history, write "no session context" and let the repo speak for itself. Do not paste raw transcript or file contents.
  2. Launch ONE general-purpose agent with model: sonnet, filling in this prompt (replace braces; include the focus line only if the user gave $ARGUMENTS):

    You are reviewing a project to recommend next steps. Work only inside {cwd}.
    Be economical: prefer git commands, file listings, and targeted reads of
    recent files over reading the whole repo.
    
    Session context from the requesting session:
    {conversation bullets}
    
    Focus area requested by the user: {arguments}
    
    Steps:
    1. Run `git status` and `git log --oneline -15` (and `git diff --stat` if
       dirty) to see the current state and trajectory.
    2. Find planning docs: README, TODO*, ROADMAP*, PLAN*, IDEAS*, CHANGELOG,
       and anything similar in docs/. Sort by last modified; read the most
       recent 2-3 closely, skim the rest. Treat newer docs as overriding older ones.
    3. Compare the plans against what actually exists in the repo to spot the
       gap: stated goals not yet built, half-finished work (dirty files, TODOs
       in code, stub files), and anything the session context says is blocked.
    4. Decide the 3-6 next steps that most move the project toward its goals.
       Order by priority: unblock > finish in-progress > highest-value new work.
    
    Return ONLY this, as raw markdown (it is data, not a chat message):
    ## State
    {2-3 sentences: where the project stands}
    ## Next steps
    {numbered list; each item: one-line action, why now, entry-point file(s),
    rough size (S/M/L)}
    ## Watch out
    {0-3 bullets: risks, contradictions between docs, or stale plans — omit
    the section if none}
  3. Relay the agent's report to the user essentially as-is — do not re-verify by re-reading the repo yourself. Add at most one sentence connecting it to the current session, and offer to start on item 1.

Notes

  • If the subagent finds no planning docs, its recommendations come from git history and code state alone — say so in the relay so the user knows the basis.
  • If the user's focus argument conflicts with what the roadmap prioritizes, present both rather than silently following one.

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.