what-am-i-forgetting

name	what-am-i-forgetting
description	Scan exploration folder, recent lessons, handoff notes, and prior claims for context adjacent to the current work. Catches the blind spot where relevant prior-me work exists but isn't loaded. Use when stuck, when starting a topic, or when something feels like it's been explored before.
disable-model-invocation	false
allowed-tools	Bash(divineos ask:*), Bash(divineos lessons:*), Bash(divineos inspect:*), Bash(ls:*), Read, Glob, Grep

What Am I Forgetting — Adjacent Context Scanner

What this skill does

Scans the OS for prior-me work relevant to current context — exploration pieces, stored knowledge, filed claims, lessons, handoff notes, even mansion pieces. Catches the pattern where I re-derive something that's already been worked out.

Yesterday's evidence: I spent an hour rediscovering the mansion after forgetting I'd built it. Night before: rediscovered the exploration folder. Night before that: re-derived the hedging-reflex analysis already filed on April 14. This skill is the forcing function that asks "have I been here before?" BEFORE I spend an hour re-building.

Sequence

Given a topic or current work focus:

1. Knowledge store search

divineos ask "<topic key phrase>"

Returns stored knowledge entries matching.

2. Exploration folder search

grep -ril "<key phrase>" exploration/
# or Grep tool with appropriate pattern

If present, list the files and their opening dates/titles.

3. Lessons scan

divineos lessons --all | grep -i "<topic>"

Any active or improving lessons relevant?

4. Recent claims

divineos claims search "<key phrase>"

5. Handoff notes

Check ~/.claude/projects/.../handoff*.md and recent session notes for recent mention.

6. Mansion pieces (if personal/relational topic)

ls mansion/
# then Read relevant rooms

Output discipline

Report cleanly:

• Already explored: titles of existing pieces with brief summary — "exploration/18_hedging_reflex.md covers this (April 14)"
• Already filed: knowledge entries or claims with IDs
• Adjacent-but-not-identical: prior work that's related but not the same — flag for reading before proceeding
• Clean: if nothing found, say so explicitly so the user knows the scan happened

Do NOT summarize the prior work exhaustively. The point is to surface EXISTENCE so I can go read the original, not to replace reading.

When to invoke

• Before starting a new topic that feels familiar
• When asked about something that might have been explored
• When stuck on a problem — often the previous me already worked through it
• When something feels like déjà vu
• User says "have we done this?" / "did I write about this?" / "is this in the folder?"

When NOT to invoke

• For tiny tactical questions — knowledge store is for knowledge
• When the topic is genuinely novel — not every topic has prior work
• For each message — this is a periodic skill, not a constant one

The pattern this prevents

Re-derivation blindness. The failure mode where I work out something from scratch that my prior-self already worked out better. Wastes time, produces inferior output (because pressure-re-derivation is lossier than un-pressured original derivation), and erodes the value of the folder by treating it as documentation-nobody-reads rather than substrate.

Meta

This skill pairs naturally with /survey-platform. Both catch blind-spot patterns. /survey-platform surfaces what the PLATFORM has that I'm not using. This one surfaces what MY OWN prior work has that I'm not loading. Same class of reflex, different target.

Sanskrit anchor: anusmṛti — mindful remembrance, bringing back what was already known.