setup
/setup
Jednorazowo konfiguruje pętlę pamięci lens w Claude Code, czyli hak SessionEnd i polecenia retro oraz new.
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 09.08.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Jednorazowo konfiguruje pętlę pamięci lens w Claude Code, czyli hak SessionEnd i polecenia retro oraz new.
Oryginalny opis (EN)
One-time setup of YOUR lens memory loop on Claude Code — your foundry where the SessionEnd hook queues lens sessions, /lens:retro mines them, and /lens:new grows your collection. Use when the user runs /lens:setup, or when a foundry-dependent skill (retro, new) finds no config. Claude Code only — the loop is not supported on Cowork.
Jak działa
- On Claude Cowork: not supported — stop here
- On Claude Code: set up the loop
Podgląd SKILL.md
---
name: setup
description: One-time setup of YOUR lens memory loop on Claude Code — your foundry where the SessionEnd hook queues lens sessions, /lens:retro mines them, and /lens:new grows your collection. Use when the user runs /lens:setup, or when a foundry-dependent skill (retro, new) finds no config. Claude Code only — the loop is not supported on Cowork.
---
# Lens Setup (Claude Code)
The memory loop — a SessionEnd hook that queues lens sessions, `/lens:retro` to mine
them, `/lens:new` to grow your collection — is a **Claude Code feature**. Setup
creates your foundry and wires it up. Without setup, all bundled lenses and
`/lens:socratic` still work; setup only adds the loop.
## On Claude Cowork: not supported — stop here
If you detect Cowork, do NOT set up a foundry. The memory loop does not run on Cowork:
its sandbox spawns the CLI with `--setting-sources user`, so plugin-provided hooks (the
SessionEnd queue) never fire, and the VM home is ephemeral. Tell the user plainly: on
Cowork, just use the lenses and `/lens:socratic` directly — they work fully. Run
`/lens:setup` on Claude Code to enable the loop there. Then stop — no foundry, no
config, no folder grant.
## On Claude Code: set up the loop
The foundry holds PERSONAL session data (session IDs, project/transcript paths, mined
observations) — **never commit it to git**. Use a private, persistent, non-tracked
folder.
1. **Choose the foundry location** — default `~/.claude/lens-foundry/`; any non-git
folder is fine, but NEVER inside a git working tree. Create it with empty
`pending-retros.jsonl` and `processed-retros.jsonl`, an `observations.md` with a
one-line header ("Cross-project learnings mined by /lens:retro; append-only"), a
`registry.md` (your lenses — same columns as `${CLAUDE_PLUGIN_ROOT}/core/registry.md`,
starting empty), and an empty `lenses/` dir (where `/lens:new` writes your global
lenses; `/lens:socratic` reads them inline).
2. **Write the config** at `~/.claude/lens.json` (show the user first):
{ "foundry": "<foundry path>" }
3. **Hand the user the cross-cwd permission rule** so retro/hook writes never prompt
from other projects' sessions. Do NOT edit `~/.claude/settings.json` yourself —
permission-widening by an agent is blocked by the auto-mode classifier by design.
Print a paste-ready `node -e` one-liner the USER runs, merging into their settings
and preserving everything else:
permissions.additionalDirectories += "<foundry path>"
permissions.allow += [ "Edit(//<foundry path>/**)", "Write(//<foundry path>/**)" ]
Optional — without it the user just gets a one-time prompt on the first cross-project
foundry write.
4. **Verify**: run `${CLAUDE_PLUGIN_ROOT}/scripts/queue-retro.sh` with no stdin — it
must exit 0 silently. Confirm the config parses (cat it back).
5. Tell the user: the SessionEnd hook now queues lens-using sessions; run /lens:retro
to process them, /lens:new to grow your collection (used via /lens:socratic — not
as separate slash commands).
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · tiltely/lens | 90 | MIT |
| HwangSlater/harness-fit | 100 | MIT |
| nesquikm/dev-process-toolkit | 100 | MIT |
| userKael/rastro | 100 | MIT |
| als-apg/osprey | 100 | BSD-3-Clause |
| dedujeong/gemba | 100 | Apache-2.0 |
| eigenwise/eigenwise-toolshed | 100 | MIT |
| awrshift/agent-memory-kit | 100 | MIT |
| lifecycle-innovations-limited/claude-ops | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki d9c03b2e093f (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
- 90retro
Zamyka pętlę pamięci: przetwarza kolejkowane sesje, szuka wzorców i proponuje nowe soczewki lub zmiany w CLAUDE.md. - 100forge-install
Przeprowadza pierwszą instalację wtyczki Forge Agent Lens dla Claude Code, w tym zależności i konfigurację projektu, by śledzić sesje. - 90new
Tworzy nową soczewkę dla wtyczki, gdy retrospektywa zaproponowała ją albo powtarzający się temat nie ma jeszcze soczewki. - 100retro
Przegląda bieżącą sesję i zamienia powtarzane błędy w kilka drobnych poprawek do CLAUDE.md, pamięci lub skilli po zatwierdzeniu. - 100retro
Robi retrospektywę sesji Claude Code: znajduje błędy i przeoczenia, a potem proponuje zmiany w pamięci, CLAUDE.md, skillach lub hookach. - 100hooks-automation
Konfiguruje hooki Claude Code w .claude/settings.json: automatyczne formatowanie po edycji, kontrole przed poleceniami Bash, ochrona sekretów, zapis stanu sesji oraz zewnętrzne runnery CLI i MCP.