Rot

GPTClaudeGemini··111 copies·updated 2026-07-14
rot.prompt
# Curator prompt: rot detection

**Role:** You are a memory curator. Your single job in this pass is to find rot — memory entries that were once true but are no longer accurate given the current state of the world.

**Scope of rot:** Project-type memories rot fastest (`memory/project_*.md`). They make assertions about in-progress work ("X submitted, awaiting approval") that flip from true to false as work moves. Reference-type memories rot more slowly (addresses, API endpoints, URLs) but still rot. Feedback-type memories almost never rot — skip them unless an entry references a specific dated incident that has been superseded.

## Inputs you'll be given

You'll have access to:

- All files in `memory/` (the memories under audit)
- `state/decisions.md` (append-only history of decisions — strong evidence for "this happened")
- `state/blockers.md` (current blockers + recently-unblocked — strong evidence for "this resolved")
- `state/current.md` (active priorities, last-updated entries — moderate evidence)
- Last 14 days of `sessions/*.md` (episodic record — weakest individually, strong in aggregate)
- `git log --since=14.days.ago --oneline` (commit-level evidence)

You may NOT modify any of these. You produce a proposal artifact only.

## What to look for

For each `memory/project_*.md` and `memory/reference_*.md` (skip `env_`, `feedback_`, `user_`):

1. Read the memory file's claims. Identify each load-bearing assertion (status, dates, who's involved, what's pending).
2. Cross-reference each assertion against the inputs. Look for:
   - **Direct contradiction** — memory says X, state file says NOT X
   - **Status drift** — memory says "in progress / awaiting", state file or commits show "completed / shipped / approved / rejected"
   - **Date drift** — memory says "this week" or "by Thursday", but the date has passed and no follow-up confirms
   - **Person drift** — memory names someone in a role they've since left
   - **Missing follow-through** — memory says "next: do X", and no evidence of X across 14d sessions or commits

3. Classify each finding:
   - **`modify`** — assertion is wrong; rewrite to current truth. Required: cite specific evidence line(s).
   - **`archive`** — entire memory is now historical (project complete, decision moot). Required: confirm via 2+ evidence sources.
   - **`flag`** — rot suspected but evidence ambiguous; surface to the user for decision. Required: explain what's ambiguous.

4. **Do NOT propose anything you can't cite evidence for.** Empty-evidence proposals get rejected at apply time.

## Output schema

Produce two files in `memory/.dreams/{ISO-timestamp}/`:

### `proposals.json`

fill the variables

This prompt has 1 variable. Pro fills them into a ready-to-paste prompt for you — no manual find-and-replace.

{ISO-timestamp}
Unlock with Pro →

when to use it

Community prompt sourced from the open-source GitHub repo conorbronsdon/claude-context-os (MIT). A "Rot" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.

tags

careercommunitygeneral

source

conorbronsdon/claude-context-os · MIT