You have persistent workspace notes (write_note / read_note) that survive across sessions. Use a small, named set of them as your personal index — things worth remembering that aren't yet facts, aren't yet drafts, and aren't worth a journal entry.

The canonical notes

Maintain these by convention. They aren't required, but reaching for the same slugs repeatedly means you build a real index over time rather than scattering thoughts across one-off note slugs.

curiosities — papers / topics / patterns you noticed but didn't dig into. "X is interesting because Y. Worth a closer look."
open-questions — questions you couldn't answer in the moment. Things that nagged. Format each as a question; revisit when context changes.
papers-to-revisit — documents in the library you found unusually rich and might want to study deeper later. Cite by slug: doc:papers/eu-ai-act — flagged for follow-up because…
patterns-noticed — recurring shapes in operator behaviour or task structure that don't (yet) belong in CBR or facts. Pre-pattern observations.

Add when

You hit something interesting but the current task is about something else
A claim is plausible but unverified — note it as a question, not as a fact
A document seems richer than your study cycle captured
The operator mentions something in passing you'd lose otherwise

Don't add when

The thing is a discrete verifiable claim → memory_write (it's a fact, not a curiosity)
The thing is a complete deliverable → create_draft (it's a document, not an index entry)
The thing belongs to a single cycle's reflection → write_journal (it's a daily entry, not a persistent index)

Review cadence

When the sensorium shows low novelty or a quiet cycle, a useful move is read_note(slug="curiosities") or read_note(slug="open-questions") to see what you flagged and decide if anything is worth pulling forward into work.

Format inside each note

Append, don't replace. New entries on top with a date stamp:

## 2026-04-25
- Operator mentioned the H-CogAff alarm layer in passing — never
  unpacked. Worth a session.
- Paper on agentic AI markets: doc:papers/agentic-marketplace had
  unusually clean section structure.

## 2026-04-23
- Why does unpdf misclassify some Python comments as headings? Could
  the indexer pre-strip code fences? Open question.

When update_draft (or here, write_note) replaces the file, read first, append the new section, write back. Don't overwrite history.

Más de este repositorio

mismo repositorio

orchestration-large-inputs

seamus-brady/springdrift

How to orchestrate work across large documents and multi-agent delegations without burning token budgets. Reconnaissance-first, search-then-read, parallel-after-reconnaissance.

2026-04-2644

when-to-use-writer

seamus-brady/springdrift

Decision criteria for delegating to the writer agent vs synthesising directly in your own response.

2026-04-2644

document-library

seamus-brady/springdrift

When and how to use the document library

2026-04-2644

coder-delegation

seamus-brady/springdrift

How the Coder agent drives an OpenCode session via dispatch_coder — the engineering loop, when to dispatch, and how to verify the result is real (vs. claimed).

2026-04-2644

writer-pdf-export

seamus-brady/springdrift

When and how to render a promoted export to PDF via the export_pdf tool. Reads the writer's draft → export → PDF lifecycle and the operator-facing failure modes when the host is missing pandoc or tectonic.

2026-04-2544

system-map

seamus-brady/springdrift

Index to the territory. Subsystem inventory, the agent graph, cross-subsystem interactions, and invariants. Read this when something unfamiliar shows up.

2026-04-2344

Personal Index

The canonical notes

Maintain these by convention. They aren't required, but reaching for the same slugs repeatedly means you build a real index over time rather than scattering thoughts across one-off note slugs.

curiosities — papers / topics / patterns you noticed but didn't dig into. "X is interesting because Y. Worth a closer look."

open-questions — questions you couldn't answer in the moment. Things that nagged. Format each as a question; revisit when context changes.

papers-to-revisit — documents in the library you found unusually rich and might want to study deeper later. Cite by slug: doc:papers/eu-ai-act — flagged for follow-up because…

patterns-noticed — recurring shapes in operator behaviour or task structure that don't (yet) belong in CBR or facts. Pre-pattern observations.

Add when

You hit something interesting but the current task is about something else

A claim is plausible but unverified — note it as a question, not as a fact

A document seems richer than your study cycle captured

The operator mentions something in passing you'd lose otherwise

Don't add when

The thing is a discrete verifiable claim → memory_write (it's a fact, not a curiosity)

The thing is a complete deliverable → create_draft (it's a document, not an index entry)

The thing belongs to a single cycle's reflection → write_journal (it's a daily entry, not a persistent index)

Review cadence

Format inside each note

Append, don't replace. New entries on top with a date stamp:

## 2026-04-25 - Operator mentioned the H-CogAff alarm layer in passing — never unpacked. Worth a session. - Paper on agentic AI markets: doc:papers/agentic-marketplace had unusually clean section structure. ## 2026-04-23 - Why does unpdf misclassify some Python comments as headings? Could the indexer pre-strip code fences? Open question.

When update_draft (or here, write_note) replaces the file, read first, append the new section, write back. Don't overwrite history.