Memory Docs Documentation

August 11, 2026 · View on GitHub

Generated by fab memory-index — do not hand-edit. Descriptions come from each file's description: frontmatter.

FileDescription
distillThe /docs-distill-memory [] skill — rewrites a memory domain's topic files to the FKF §3.2/§3.3 present-truth style: strips narration, superseded state, and change-id heading suffixes; dedupes byte-identical blocks; rewrites DD changelog bullets; relocates TODOs to the backlog. One domain per approval unit, propose-then-apply. optional: named forces a full read; omitted surveys all domains then loops every flagged one sequentially. Intent-first classifier keeps durable intent.
hydrateThe /docs-hydrate-memory skill — argument routing, three modes (ingest/generate/backfill), the writer contract, index-ownership model. Ingest/generate rewrite each topic section to current truth (no change-id headings, rationale in four-field Design Decisions) then self-check before regen; backfill is body-preserving and self-check-exempt. Refuse-before-regen guard keys on fab memory-index --check exit 2, distinct from the exit-1 blocking class (malformed/change-id/over-cap description).
hydrate-generate/docs-hydrate-memory generate mode — codebase scanning, gap detection, interactive scoping, memory file generation + placement rules (target path, domain/sub-domain index stubs, shape bounds); generated files carry a change-id-free description and a present-truth body (no ## Changelog)
hydrate-specs/docs-hydrate-specs skill — structural gap detection between memory and specs, interactive propose-then-apply, incl. the no-target new-spec-file branch and aligned prompt/handler tokens
specs-indexdocs/specs/ — pre-implementation specs, specs-vs-memory distinction, bootstrap/context integration; per-skill Flow skeletons live in docs/specs/skills.md (one per skill section, behavioral partials under § Skill Helpers); no specs-index generator — hand-rewritten index, no frontmatter backfill (no-symmetry note)
templatesArtifact templates (intake, plan, memory.md) and the memory file format: FKF type + change-id-free description (500-char soft cap, blocking past 1000), present-truth body style (no narration, change-id headings, or TODOs; rationale in four-field Design Decisions), generated domains-only index + C-lite/freeze-on-write log.md, no per-file changelog. Memory Tree Shape homes the shape/debt-meter warnings, the blocking content class, never-hand-merge policy, and docs-reorg-memory's rebalancer.