| name | thoughts |
| description | A durable, disposable home for in-flight thinking that overflows a single context window — layered scratch → (optional) blackboard → (optional) conclusions under thoughts/<task-slug>/, refuted by the existing adversarial-verify and drained into durable truth (scribe / deep-analysis) when it settles. Trigger: ONLY when one task won't fit in one context (multi-session work, or a deep-work/deep-analysis run whose shared board must survive the window). Below that bar, think in-context.
|
Thoughts — externalize only when one head won't hold it
This is a convention, not a system. It reuses everything: the blackboard is deep-work's /
deep-analysis's existing multi-agent synthesizer surface, persisted; the membrane is
adversarial-verify; the durable exit is scribe's capture→curate→verify and docs/deep-analysis/.
The only net-new is when to persist and where.
When to externalize (the trigger — and the restraint)
Externalize only when a single task genuinely will not fit in one context window — it spans
multiple sessions, or a deep-work/deep-analysis run's shared board must outlive this window. Below
that bar, keep thinking in-context: a thoughts/ directory for a task that fits one head is ritual.
There is no "always externalize" rule, and nothing auto-creates this.
The layers (one required, two optional) — thoughts/<task-slug>/
scratch.md (required) — free-form, append-only running notes: the flipchart. One file per task,
no inner schema or template. This is where thinking lives between sessions.blackboard.md (optional) — created only when work actually fans out to agents — the
persisted form of the existing synthesizer surface, where parallel findings converge.conclusions.md (optional) — created only when something actually survives to promote — a
staging area that drains, not a destination.
A solo multi-session overflow needs only scratch.md. Don't create the others speculatively.
The membrane and the exit (both reuse existing machinery)
- Membrane: a thought moves
scratch → conclusions (and out) only after it survives
adversarial-verify — refute-first; multiple lenses only when it matters. Nothing is "concluded"
just by being written down.
- Exit (drain to durable truth): when a thought settles, a human/brain semantic act moves it
out via existing machinery — a decision/error →
docs/.scribe/capture.log (→ curate → verify into
ADRs/incidents/STATUS); an audit finding → docs/deep-analysis/. This is never an
auto-promotion hook (a decision is semantic — only the working mind knows it happened). Once drained,
thoughts/<task-slug>/ is disposable — delete it; the durable truth now lives in docs/.
Not tracked (and why)
Add /thoughts/ (root-anchored) to the project's .gitignore — justified directly by its
disposable, drainable nature: it's in-flight scratch that drains into docs/ (the tracked source
of truth) and is then deleted. Anchor it with the leading / so it ignores only the project-root
working dir, never a directory merely named thoughts elsewhere in the tree. (This is not because
docs/.scribe/ is gitignored — it isn't; the curator reads it from the repo. thoughts/ is ignored
on its own merits.)
What this does NOT build
No engine / runtime / daemon / workflow-script; no DB / schema / FTS; no new parallel router or
membrane engine; no new doc types; no hook that auto-creates or auto-promotes thoughts/; no rigid
template inside scratch.md; no 6th plugin; no "always externalize" rule. Route is untouched — this is
a brain-internal escalation of deep-work/deep-analysis when their board must persist, not a new lane
("Router = what pipeline; brain = how deep").
