<< All versions

Skill v1.0.0

currentLLM-judged scan90/100
tedhaley/pensieve/pensieve-plan
──Details
PublishedOctober 21, 2026 at 10:27 AM
Content Hashsha256:ce4cbc3f4063d5a3...
Git SHAed38747d7f2b
──Files
Files (1 file, 1.7 KB)
SKILL.md1.7 KBactive
SKILL.md · 33 lines · 1.7 KB

version: "1.0.0" name: pensieve-plan description: Ground a plan for new code in what already exists and who knows it, using Pensieve. Use when planning a feature, refactor or new module, writing a design doc, or before starting a large change.


Planning with prior art and expertise

Before proposing a design, spend a few Pensieve calls finding what already exists and who has context. Plans that reuse existing patterns and name the right people get built faster and reviewed easier.

Steps

  1. Prior art. search for the capability in plain words, and for any names you expect it to need. On the best

hits call similar(id) to find related implementations across the repo (and other indexed repos).

  1. Earlier attempts. search with kind:session for previous agent sessions on the topic: decisions, dead

ends, and why they were abandoned.

  1. Expertise. people(topic="<capability>", repo=...) for the people who wrote the closest code;

people(repo=...) to make sure that they are still active; repo(repo, part="unexplored") for nearby areas that the user has not worked in yet.

  1. Risk. repo(repo, part="gaps") shows the areas that the change touches whose authors have left.

Put it in the plan

Add these sections (keep each short):

  • Prior art: files/functions to reuse or follow, with paths; near-duplicates to avoid.
  • Earlier attempts: what was tried before and what was learned (link the session if useful).
  • People to loop in: one or two names, each with why (the code they wrote) and the question to ask them.
  • Knowledge risks: areas with a single remaining expert or none; whether to document as part of the work.

Suggest conversations; never contact anyone yourself.

All versions