feat: evict closed issues from the local store

The store is a working set, not an archive. Until now nothing removed a
closed issue from it: #10 put a filter on the write and said so explicitly
("existing store files are not cleaned"), and the migration was never
anybody's job. The only way out was rm past every script, followed by
rebuilding INDEX.md by hand.

issue_evict.py removes <id>.md and every sidecar under that slug for an
issue that is state: closed AND carries an origin: naming a tracker, then
rebuilds INDEX.md. --dry-run prints and writes nothing at all.

Two conditions, and the second one is the whole safety argument. An
origin: local issue IS the work — there is no other copy — so it is never
evicted, in any state, not even when named on the command line: it is
reported and kept. The only files that go are ones whose own metadata says
pull.py <n> brings them back, which is the trade push.py already makes
when it drops a file the tracker just confirmed.

The command lives in the domain layer, and the layering rule decides that
rather than convenience: state: and origin: are domain fields and the
answer is already on disk, so eviction needs no network, no login and no
tea. The domain also gains issue.slug_files — every file the store holds
under one slug, which is all_ids' "a slug has no dot in it" read the other
way round, and lets the domain remove an issue completely without learning
what a comment thread is.

skills/sync/scripts/evict.py is the bridge form, and it exists because a
local state: is only as fresh as the last pull: an issue closed in the web
UI still reads open here. It refreshes state: from Gitea, then calls
issue_evict.run — one implementation of "what may be evicted", in the
layer that owns the fields it reads. Same gate as push, one step earlier:
every candidate's state is fetched before anything is removed, each answer
must be an object carrying the number asked about and a state the domain
recognizes (confirmed_state, the counterpart of confirmed_number), and a
failed or unconfirmed call evicts nothing — not even the candidates whose
answers had already arrived, and no refreshed state: is written back
either. A candidate is an issue with a gitea: handle; origin: local has
none, is never asked about, and is never removed.

.remote.json is deliberately not pruned. It is the number -> slug ledger,
its entries are supposed to outlive the files they name, and an evicted
issue is in exactly the state a pushed one is.

AGENTS.md gains the rule the tracker side never wrote down: pull by number
fetches an issue in any state — an address is not a query. Eviction does
not revoke it, so a closed issue pulled after a cleanup is on disk again,
and that is the tracker answering what it was asked.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

skills/issue/SKILL.md

@@ -38,6 +38,7 @@ All offline, all in `<skill-base-dir>/scripts/`.
 | `issue_check.py [id…]` | validate against the canonical format; exit 1 on errors |
 | `issue_ac.py <id> [--check N\|TEXT]` | list the body's checkboxes; tick or untick one |
 | `issue_tree.py [id…]` | draw the dependency graph from `depends:` |
+| `issue_evict.py [id…] [--dry-run]` | remove closed issues from the store; **never** an `origin: local` one |
 | `issue_index.py` | rebuild `tmp/issues/INDEX.md` |
 | `issue.py` | the domain module the others import — not a command |
 
@@ -205,6 +206,50 @@ on `tmp/issues/<id>.md`, and this layer does not know the difference. Getting
 the rewritten body into the tracker is a separate decision — `push.py --update`
 in `/tea:sync` — and is no part of this.
 
+## Evicting closed issues
+
+The store is a working set, not an archive. A closed issue is not a unit of
+work any more, and one command takes it out — no `rm`, no rebuilding `INDEX.md`
+by hand:
+
+```bash
+python3 <skill-base-dir>/scripts/issue_evict.py --dry-run   # what would go
+python3 <skill-base-dir>/scripts/issue_evict.py             # every closed one
+python3 <skill-base-dir>/scripts/issue_evict.py old-thing   # just this one
+```
+
+Two conditions, both read off the file, and the second one is the whole safety
+argument:
+
+| `state:` | `origin:` | what eviction does |
+|---|---|---|
+| `closed` | a tracker | removes `<id>.md` and every sidecar under that slug |
+| `closed` | `local` | **keeps it, always**, and says why |
+| `open` | anything | keeps it |
+
+**`origin: local` is never evicted, in any state, not even when you name it on
+the command line.** That file *is* the issue; there is no copy to fetch back.
+Only a file whose own metadata says the work lives somewhere else may go — the
+same trade `push.py` makes when it drops a file the tracker just confirmed.
+
+- `--dry-run` prints what would go and writes nothing at all, `INDEX.md`
+  included.
+- `INDEX.md` is rebuilt afterwards, so the table and the directory agree. It is
+  rebuilt only when something was actually removed.
+- `.remote.json` is **not** pruned, deliberately: it is the number → slug
+  ledger, and its entries are supposed to outlive the files they name (that is
+  what makes `pull.py <n>` land on the same slug after a push). An evicted issue
+  is in exactly the state a pushed one is.
+- **This is not a one-off migration.** `pull.py <n>` fetches an issue in any
+  state — a number is an address, not a query — so a closed issue pulled after
+  an eviction lands on disk again. Not a regression: evict it again when you are
+  done reading it.
+
+This command is offline and decides from `state:` in the file, which is only as
+fresh as the last pull. To have the tracker's answer instead — an issue closed
+in the web UI five minutes ago — use `/tea:sync`'s `evict.py`, which refreshes
+`state:` first and then calls exactly this decision.
+
 ## Dependency graph
 
 `depends:` is the authoritative edge list; the body's `## Depends on` section