# Sks Gc

> Use when reclaiming resources leaked by shikanime jj workflows — dangling bookmarks, skill-created jj workspaces, and leftover working-copy dirs from sks-async/sks-dev-workflow.

- Skill: `shikanime-labs/sks-gc` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add shikanime-labs/sks-gc`
- Raw SKILL.md: https://api.skillmd.com/api/skills/shikanime-labs/sks-gc/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- License: Apache-2.0
- Author: shikanime-labs (https://skillmd.com/u/shikanime-labs)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/shikanime-labs/sks-gc

---


# Shikanime Org Workspace Garbage Collection

Reclaim resources leaked by the parallel/migration skills: `jj` bookmarks with
no open PR and not on `main`, `jj` workspaces the skills created
(`<repo>.<unit>`, `<repo>.fix`), and the sibling working-copy directories they
leave on disk. Destructive — always dry-run first.

## Available script

- `scripts/discover.sh [REPO_DIR]` — print every GC candidate: dangling
  bookmarks (not trunk, no open PR) and skill workspaces with a clean/dirty
  guard per workspace. Dry-run only; never forgets or removes. Run it first,
  review the list, then apply manually (steps 2–4).

## When to Use

- After a finished or stalled `sks-async` fan-out or `sks-dev-workflow`
  isolation pass: workspaces and bookmarks pile up.
- "Clean up my dangling branches / orphan jj workspaces."
- Periodic hygiene on a repo worked across many parallel streams.

## Hard rules (do not skip)

- **Dry-run first.** Print every candidate; never `forget`/`rm` on the same pass
  that discovers it. Require an explicit apply step after review.
- **Never drop `main` / `trunk` / `master`** or a bookmark with an OPEN PR.
- **Never forget a workspace with uncommitted changes.** Skip it and report —
  losing WIP is data loss.
- Operate in the repo root (`~/Source/Repos/.../<repo>`); the canonical
  workspace (named after the repo, no unit suffix) is NEVER a candidate.

## Procedure

1. **Discover** — run the dry-run scanner, review its output:

   ```bash
   cd ~/Source/Repos/github.com/<orga>/<repo>
   bash <skill-dir>/scripts/discover.sh   # scripts/ resolves against the
                                          # skill dir, not the target repo
   ```

   The script lists `== dangling bookmarks ==` (names only; trunk and
   open-PR bookmarks already excluded) and `== skill workspaces ==`
   (`CLEAN <name> <path>` vs `DIRTY <name> <path>`).

2. **Apply** (only after review) — forget dangling bookmarks and clean
   workspaces:

   ```bash
   jj bookmark forget <dangling...>
   jj workspace forget <name...>       # already confirmed clean
   rm -rf <path>                        # ONLY the sibling dir from step 1
   ```

3. **Prune remote-tracking bookmarks no longer on origin** (safe, built-in):

   ```bash
   jj git fetch --prune --remote origin
   ```

4. **Verify.**

   ```bash
   jj workspace list && jj bookmark list
   gh pr list --state open
   ```

## What "resources from other skills" covers

- `sks-async` → one `jj workspace add ../<repo>.<unit>` per stream and a
  bookmark + PR branch per link on `origin`.
- `sks-dev-workflow` → `jj workspace add ../<repo>.fix` isolation dirs and
  `/tmp/wip*` scratch copies (the latter are manual — list, but never
  auto-`rm` without asking).
- Parallel-stream chains → branch bookmarks on `origin`; cleared by forgetting
  the bookmark plus `jj git fetch --prune` (step 3).

## Pitfalls

- `jj workspace forget` does NOT delete the on-disk directory — `rm -rf` it
  yourself after confirming it is the sibling skill dir, never the repo root.
- A dirty workspace skipped by the guard means real WIP; surface it, do not
  force.
- Remote-tracking bookmarks (`origin/...`) need `--prune` (step 3), not
  `bookmark forget`, to clear.
- The canonical workspace (bare repo name, no dot/`-fix`) is never a candidate;
  don't fold the trunk working copy into GC.
- **jj 0.43 output formats.** `jj workspace list` default lines have no path
  column, and `jj bookmark list` prints `name: changeid desc` plus indented
  `@origin` continuation lines — parsing those with `awk '{print $1}'` yields
  trailing colons and `@origin` noise. `discover.sh` templates the exact fields
  (`-T 'name ++ "\t" ++ root'` for workspaces, `-T 'name ++ "\n"'` for
  bookmarks); don't revert to awk on the default output.

## Verification

```bash
jj workspace list && jj bookmark list
gh pr list --state open   # confirm no protected bm forgotten, canonical wc untouched
```

## See also

- `sks-async` — the fan-out that creates the workspaces/bookmarks this skill
  reclaims.
- `sks-dev-workflow` — isolation workspace pattern (`../<repo>.fix`).
- `sks-land` — lands PRs (this skill only reclaims after landing, never
  merges).