Przejdź do treści
Biblioteka skilli
AI i agenci · Przegląd i audyt

memory-audit

/memory-audit

Sprawdza MEMORY.md pod kątem za długich sekcji, wzorców do przeniesienia, przestarzałych wpisów i zdublowanych liczb, a potem proponuje plan przeniesień.

Licencja
MIT
Ostatnia zmiana repozytorium
13.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Sprawdza MEMORY.md pod kątem za długich sekcji, wzorców do przeniesienia, przestarzałych wpisów i zdublowanych liczb, a potem proponuje plan przeniesień.

Wyzwalacze z opisu

audit memoryпроверь памятьпочисти память
Oryginalny opis (EN)

Audit MEMORY.md against the memory discipline — oversized sections, settled multi-session patterns that belong in knowledge/concepts/, session-headed chronicle blocks, restated rules, copied numbers, stale entries. Produces a move plan as a table for approval, then executes the approved moves atomically. Use when the SessionStart hook reports a tripped cap or session-headed blocks, when PreCompact blocks on an oversized cache, or when the user says "/memory-kit:memory-audit", "audit memory", "проверь память", "почисти память". Refuses only when no cap is tripped, no session block is flagged AN

Jak działa

  1. Inputs
  2. Step 1 — classify every section
  3. Step 2 — the move plan (a table, never prose)
  4. Step 3 — PAUSE
  5. Step 4 — execute the approved moves, atomically
  6. Step 5 — the header, then the report
  7. Refuse when
  8. Anti-patterns
Sekcje z nagłówków SKILL.md, w oryginale · 96 linii, 1 054 słów
Podgląd SKILL.md
---
name: memory-audit
description: Audit MEMORY.md against the memory discipline — oversized sections, settled multi-session patterns that belong in knowledge/concepts/, session-headed chronicle blocks, restated rules, copied numbers, stale entries. Produces a move plan as a table for approval, then executes the approved moves atomically. Use when the SessionStart hook reports a tripped cap or session-headed blocks, when PreCompact blocks on an oversized cache, or when the user says "/memory-kit:memory-audit", "audit memory", "проверь память", "почисти память". Refuses only when no cap is tripped, no session block is flagged AND no settled-pattern candidate exists.
allowed-tools: Read, Grep, Glob, Edit(.claude/memory/**), Edit(knowledge/**)
---

# Memory audit — the surgical one-file pass

This is the **cap-trip response**, not the daily ritual. `/memory-kit:close-session` runs every
session and captures; this runs when the hot cache has actually outgrown itself and something
must LEAVE it. Keeping them separate is the point: a daily ritual that also has to do surgery
becomes a ritual people skip.

**Always pause for approval before writing.** The user talks; you write — including here.

## Inputs

- `.claude/memory/MEMORY.md` — read in full.
- `knowledge/concepts/*.md` — filenames + first 5 lines, for duplicate detection (the directory
  may not exist yet; the first move creates it).
- `context/handoffs/` — to confirm a session narrative already lives there → **drop**, not move.
- The newest `context/audits/memory-audit-deferred-*.md`, if one exists — the candidates the
  previous audit left; start from them.

## Step 1 — classify every section

Scan by `## ` heading (and `###` inside a long section). For each, count lines and mark exactly
one action. Also read the project's `CLAUDE.md`, the `.claude/rules/*.md` without `paths:` and
the plugin's `reference/` titles — you need them for `restates-rule`.

| Mark | When |
|---|---|
| `move (create)` | a settled pattern (confirmed on 3+ distinct dates) with >25 lines of reference detail |
| `move (merge)` | same, but a concept article on the topic already exists |
| `session-block` | the heading names a session or a date, not a topic (`### s66 (2026-09-25)`, `## Session 65 wrap`, `## Findings [date] — …`; the SessionStart hook flags these) → dissolve by topic: each settled lesson → one line under its topic heading or a concept; the rest → drop |
| `drop` | a per-session chronicle — it already lives in a handoff; distil at most one settled line |
| `restates-rule` | the entry says what `CLAUDE.md`, an always-loaded rule or a `reference/` file already says → drop, or one pointer line if the link is not obvious. The agent reads those files every session; a second copy only drifts |
| `point to SSOT` | a derivable number — price, count, version, limit — whose source of truth lives elsewhere (a price table in code, a config, a vendor page) → replace the value with a pointer to that source (`reference/doc-governance.md` R1). A copied number outlives the thing it described |
| `update in-place` | stale: names a deleted file, a closed ticket, a superseded decision |
| `simplify` | true and useful, but three times longer than it needs to be |
| `keep` | earns its lines |

## Step 2 — the move plan (a table, never prose)

| # | Section | Lines | Action | Target | Reason |
|---|---|---|---|---|---|
| 1 | … | N | move / merge / session-block / drop / restates-rule / point to SSOT / simplify / keep | `knowledge/concepts/<slug>.md` or — | … |

Close the table with: estimated line savings, and the projected line/byte count afterwards
against the caps (180 lines / 32 KB / 3000 chars per line). **Target well below the caps — at
most ~60 % of the byte cap (≈ 20 KB).** A cache audited to «just under» refills past the cap
within days: the cap works as a fill line, not a ceiling. If the plan does not reach the target,
say which sections keep it above and why they earn their bytes.

## Step 3 — PAUSE

Print the plan and ask for approval. Wait for an explicit yes or edits. No writes before that.

## Step 4 — execute the approved moves, atomically

- **move (create):** write `knowledge/concepts/<slug>.md` with frontmatter matching
  `knowledge/index.md`'s spec, then add its one-line entry to that index in the same pass.
- **move (merge):** put the fact INTO the body of the article's topic section, with a short date tag
  (`[since YYYY-MM-DD]`); the line it replaces is deleted — no `## History` section, the past lives in git.
  Never append a dated «From MEMORY.md» block at the end — appended blocks turn an article into a log nobody can
  read as current truth. Never duplicate a section that already says the same thing.
- **in MEMORY.md:** replace the moved section with a one-line pointer plus a one-line teaser.
  A move without a pointer orphans the content — that is the failure this audit exists to prevent.
- **drop:** delete; keep at most one distilled line.
- **session-block:** put each surviving lesson under its topic heading (create the topic heading
  if none fits), then delete the session heading and the rest of its block.
- **restates-rule / point to SSOT:** delete the copy; where a reader would not find the source
  unaided, leave one line naming the file (and the symbol or section) that holds it.

## Step 5 — the header, then the report

Replace the MEMORY.md current-state header (≤3 lines, dated — or the one pointer line to the newest
handoff, when the project's session-close document sets that form). Report before/after line and byte
counts, every file touched, and save declined candidates to
`context/audits/memory-audit-deferred-YYYY-MM-DD.md` (create the dir if absent) so the next audit
starts from them. Not in `context/handoffs/`: the SessionStart hook injects the newest file there
as the session handoff, so a deferred list saved there replaces the real handoff at the next start.

## Refuse when

- No cap is tripped, no session-headed block is flagged AND no settled-pattern candidate exists →
  say "no audit needed" and stop. A SINGLE tripped cap or flagged block is reason enough to
  proceed (one 3000-char line qualifies).
- The user declines the plan → save it to `context/audits/` as above and stop.

## Anti-patterns

- Writing anything before the approval.
- Creating a new concept when a keyword grep matches an existing one — merge bias broken.
- Promoting a session CHRONICLE into a knowledge article. Concepts hold settled patterns;
  narrative belongs to handoffs, and the action for it here is `drop`.
- Stopping at «just under the cap». The next few sessions refill it; aim for the ~60 % target.
- Keeping a rule «as a reminder». If the always-loaded layer already says it, memory repeating
  it is how two versions of one rule start to disagree.
- Burying the plan in prose. The table is what makes it reviewable in ten seconds.

Wersje

RepozytoriumWynikLicencja
Ta strona · awrshift/agent-memory-kit100MIT
fastled/fastled100MIT
awrshift/claude-memory-kit100MIT
chuckplayer/claude-agent-pack100MIT
aendrix03/graft100Apache-2.0
kalabint/llm-tooling100AGPL-3.0

Migawka bazy z 13.11.2026 · wersja paczki b12768940847 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100memory-clean
    Audytuje katalog pamięci pod kątem problemów strukturalnych (osierocone i zerwane odwołania, duplikaty, brakujące sekcje, zbyt duże wpisy) i nieaktualności względem transkryptów, najpierw raportując, a poprawki po potwierdzeniu.
  • 100memory
    Pokazuje i zarządza tym, co agent pamięta między rozmowami i które pliki instrukcji wczytuje, usuwa wpisy na żądanie i sprząta.
  • 100memory-kit
    Zarządza trwałym kontekstem sesji Claude Code przez MEMORY.md: zapis, ładowanie, aktualizacja, udostępnianie i audyt pamięci, co zapobiega utracie kontekstu przy kompaktowaniu lub restarcie sesji.
  • 90memory-clean
    Przegląda pamięć agenta pod kątem nieaktualnych i zduplikowanych wpisów na prośbę człowieka, bez zapisywania nowych.
  • 100memory-pass
    Sprawdza aktualność wpisów o projektach w MEMORY.md na tle logów, klientów i decyzji po dłuższej przerwie lub zmianie statusu, nie ruszając reguł zachowania.
  • 100memory-hygiene
    Utrzymuje automatyczną pamięć Claude Code w małym, aktualnym i wiarygodnym stanie: sprzątanie, audyt i aktualizację wpisów MEMORY.md.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.