1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
# gitctl TODO
Planned features. Each note captures the non-obvious bits so the work does not
re-discover them. Keep the hard rules from CLAUDE.md (stdlib only, single-file
scripts, single-writer repo.desc, idempotent, no em dashes, work on master).
## 1. delete repo
`gitctl repo delete <name>` - remove a repo from all three places it lives.
Touches three writers, so it is the inverse of `repo create`:
- gitolite.conf stanza -> client edits the admin clone, pushes (needs the SSH
key / card, like create).
- cgit block in /etc/cgitrc -> new helper verb, the inverse of
`insert_repo_block`: find the repo's `repo.url=` line and delete its block up
to the next `repo.url=`, `section=`, banner, or EOF. Back up first
(write_cgitrc_lines already does). Run sync after.
- the bare repo on disk -> gitolite does NOT delete it when the stanza goes
away. The helper must `rm -rf REPO_BASE/<name>.git` (validate_url first;
guard against empty/`..` - validate_url already does, but double check the
final path is under REPO_BASE before rm).
Design notes:
- DESTRUCTIVE. Always confirm, show what will be removed. The `-y` flag from
item 3 must NOT bypass the delete confirmation (or gate it behind a separate,
louder opt-in). Deleting data is not the same as auto-confirming an additive
conf diff.
- Idempotent-ish: if the repo is already gone from a given place, skip that
step and report, do not error.
- No undo. The cgitrc backup covers the cgit block; the bare repo is gone for
good once rm runs. Consider moving to a trash dir instead of rm -rf, or at
least refusing if the repo has refs unless `--force`.
- Order: remove cgit block + bare repo via helper first (reversible-ish), then
the gitolite stanza push? Or stanza first? Decide and document - partial
failure (push fails after rm) should leave a clear, resumable state like
create does.
## 2. list repos
`gitctl repo list` - show repos known to the server, ideally with their section.
- New helper verb `list-repos`: parse /etc/cgitrc for `repo.url=` lines and the
preceding `section=` (reuse `find_repo_section` / the section-tracking loop).
Print `<section>\t<url>` or similar.
- Cheap, read-only, no confirmation. Good first item - unblocks better bash
completion (complete repo names for `desc` / `delete` / `add-remote`) and
gives `delete` something to validate against.
- Alternative source: gitolite (`ssh git_push info` lists repos) but cgitrc is
the helper's own domain and already parsed; prefer it.
## 3. `-y` / `--yes` flag on `repo create`
Let an agent drive `repo create` without the interactive y/N, so the only human
step left is supplying the SSH-key password / smartcard PIN for the push.
- Add `--yes` (`-y`) to `repo create`; when set, skip the `_confirm` prompt and
proceed to commit + push.
- IMPORTANT: `-y` does NOT remove the need for a human at the SSH key. The
phase-1 `git push` to gitolite-admin still requires the key's password or the
card touch/PIN, which an agent cannot supply. `-y` removes only the y/N
diff-confirmation; the push auth is still interactive. Document this clearly
so nobody expects fully unattended creation.
- Still print the diff before pushing even with `-y` (visibility), just do not
block on it.
- Consider whether `-y` should also auto-create a missing section (it already
does that without asking) - keep behavior consistent.
## Cross-cutting
- Bash completion: once `repo list` exists, complete repo names for `desc`,
`delete`, `add-remote` (like `--section` is completed today). One ssh call,
only when completing a repo-name argument.
- Skill: update both SKILL.md files (repo skills/gitctl + ~/.claude) to mention
delete (and its destructive, confirm-always nature) and `-y`'s real limit.
- Tests: each new helper verb gets self-test coverage (cgit block removal is the
risky one - mirror the insert tests: first/middle/last/missing).
|
