Six findings from the whole-branch review, all of which end in the same
place: a backup tool reporting success while nothing was backed up.
Slug migration nested the old entry inside the new one. The clear before
the rename was gated on the destination's claude/memory subdirectory
rather than on the destination itself, so an entry that exists without
one survived, `git mv A B` moved A *inside* B, and the mkdir below
fabricated a fresh empty memory directory for the live link to point at.
The real memory ended up one level deeper than --status and --restore
ever look, and the run returned 0. That shape is not exotic: git cannot
track an empty directory, so an entry committed while its memory was
empty comes back from a clone as projects/<slug>/origin and nothing
else -- and cloning the vault is this feature's own recovery path. The
destination is now moved aside the way --adopt already does it rather
than deleted (widening the rm -rf would have destroyed the clone's
origin log), its provenance is folded into the migrated entry, and every
failure path rolls back and reports.
The launch path pulled over the network. Both wrappers call agents-vault
synchronously before starting an agent, and the pull in the shared sync
helper was unguarded once an upstream existed: against a blackholed
remote it blocked the launch indefinitely and then aborted the commit,
so an offline laptop silently stopped being backed up at all. Committing
never needed a remote, so the pull moved to the push path, which was
already opt-in for exactly this reason. A failure there now distinguishes
a real rebase conflict (rebase-merge/ or rebase-apply/ present) from an
unreachable remote instead of calling both a conflict, and both network
calls set GIT_TERMINAL_PROMPT=0 and GIT_ASKPASS so they fail fast rather
than prompt with nobody watching. The helper still refuses to commit a
rebase in progress, and leaves it standing rather than aborting one it
did not start. This also restores agents-init's pre-refactor ability to
commit while offline.
The agy knowledge copy was unfiltered. The allowlist held at the agy root
and nowhere below it, so a planted .credentials.json inside knowledge/
was committed verbatim while the documentation promised nothing new
upstream added could leak in. Only *.md and *.json are copied now --
which is what the store actually holds -- so lock files, transcripts and
conversation databases are excluded by having no business in a backup
rather than by being known about. The scaffolded .gitignore also ignored
only the SQLite sidecars and not the databases, which is worse than
ignoring neither: a torn database landed in history with the write-ahead
log that would have completed it deliberately excluded. Both changes are
template-only, on a feature that has never shipped.
agents-init reported success when nothing was committed. It ended on a
branchless `if` with no arm for a failed commit, which fish resolves to
0 -- the same false zero already fixed in agents-vault, left in the
function the refactor was rewriting. It now has the arm and an explicit
final status.
The --adopt forward-failure path with no stash left a raw coreutils `mv:`
line and no statement that the adopt had been abandoned cleanly; it is
branded like every other error exit in the function.
Tests: the suite now clones a vault with git and runs agents-vault
against the clone, instead of trusting hand-built fixtures to have shapes
git can actually produce -- that blind spot shipped both of the merge
blockers. The "present but empty" migration fixture is rebuilt as the
origin-only directory a clone leaves behind, with the hand-built shape
kept as a separate case. Reverting each fix drops the suite from 285 to
279 (migration), 261 (network), 275 (knowledge allowlist) and 283
(agents-init status).
Both wrappers stay behind the C1 guard, so disabling
__fish_config_op_aliases still passes straight through to the real binary.
Launch commits but never pushes, keeping the network and any credential
prompt off the critical path; pushing is left to the Claude Code SessionEnd
hook. agy has no such hook, so its memory lands one launch later.
The adopt rollback restored the worktree but not the index. Every move it
makes is a plain rename as far as git is concerned -- the stash move out
from under the index most of all -- so a rolled-back adopt left a
half-applied rename staged against a clean vault. No bytes were at risk
and the next ordinary run's `git add -A` healed it, but a hand
`git commit` in that window recorded the half-applied state. Both
rollback sites now re-read projects/ once the worktree is whole again,
the stash restore included. projects/ is named whole rather than the two
entries, because `git add` refuses a pathspec that matches nothing --
which one of the two always is, once it has been moved back -- and then
stages neither.
The stash itself moves from the vault root into .git/, where neither the
entry walk nor `git add -A` can reach it, so a crash between the two
moves can no longer leave junk at the vault root for the next run to
commit. A vault whose .git is not a directory falls back to the root,
which the scaffolded .gitignore now covers.
--status and --restore walked the vault with a fish glob, which does not
match dot-led names. A dot-led slug is both reachable and sanctioned: the
sibling-bare-mirror idiom (`git remote add origin ../mirror.git`) keys as
..-mirror, a dot-led host keys as .hidden.example.com-o-r, and --adopt
accepts a leading dot on purpose. Such an entry is scaffolded, linked,
committed and pushed normally, yet --status under-reported it and batch
--restore left that project unlinked, both without saying so. Both walks
now list the directory instead. The --adopt completion gains -A for the
same reason: an entry that cannot be completed reads as one that is not
there.
The now-fatal push failure is painted as an error rather than a warning,
matching its sibling on the commit path.
The header notes that --adopt does not pin a name. The slug is re-derived
on every run, so the next ordinary run migrates the adopted entry back to
the canonical key, memory and live link following. Behaviour unchanged;
only the documentation gap is closed.
Tests, 173 -> 209 checks. The whole stash branch of --adopt was
uncovered, because the existing atomicity test adopts onto a slug with no
entry at all: a successful stash-adopt and a stash-adopt whose relink
fails are both pinned now, the latter asserting an empty
`git status --porcelain` and a still-reachable live memory. The stash
location is pinned by making the vault root unwritable for the duration,
which only a stash at the root would need. agents-vault's own propagation
of a failed sync had no test at all -- the third recurrence of fish's
branchless-`if` false zero here -- so both ways it can fail are now
driven end to end: a rejecting pre-commit at the vault's own
core.hooksPath, and a real rebase conflict against a bare remote. A
dot-led entry is asserted in --status and in --restore.
A push that fails against a configured remote warned on stderr and then
fell through to the trailing branchless `if`, which resolves to 0, so
`--push` reported a successful backup while nothing had left the machine.
That is the exact loss the vault exists to prevent. It now returns
non-zero, verified against a real unreachable remote rather than a mock.
The same audit found two more false zeros in this function, both fixed:
the commit block warned about a rebase conflict and walked past it, and
swallowed a hook-rejected commit entirely (neither branch of its if/else
if matched, since the error goes to stderr rather than stdout); and
--restore reported a relink failure and then returned 0 regardless. All
three now feed one flag and the function ends on an explicit status
rather than on whatever the last branchless `if` left behind.
--adopt is now atomic. A rename that landed while the relink failed left
the memory intact at the new slug but unreferenced: the next ordinary run
found no live link, recomputed the old slug, found nothing there, and
fabricated a fresh empty entry, so the agent wrote history-less memory
from then on. No bytes were lost, but continuity was, with no automated
recovery. The live link is no longer removed first -- ensure_symlink
repins a link that points elsewhere on its own -- a contentless target
entry is moved aside rather than deleted, the origin note is appended
only after the relink succeeds, and a failed relink rolls the rename back
so the vault is exactly as it was.
The --adopt validator no longer refuses a leading dot. _agents_repo_slug
legitimately emits one for a dot-led subdomain, so refusing it made such
an entry impossible to adopt; inside projects/ it is a hidden directory,
not an escape. The traversal cases are still refused: no slash survives
the charset, and "." and ".." are refused by name.
Adds the RETURNS section the header was missing. --status prints a
structured report, which this repo's convention treats as return value
rather than as progress output.
--status reports link health, orphaned entries, and how far the vault is
ahead of its remote, which is how an unpushed backup gets noticed. --adopt
rebinds a machine-specific local-* entry by hand. --push is explicit;
autopush stays opt-in via __fish_agent_vault_autopush.
Three corrections to the planned shape:
--status is dispatched ahead of the scaffold instead of behind it. As
planned it sat after the tool install, the agy knowledge copy, and the
global memory link, so asking for a report would first sync global state
and claim ~/.claude/memory. It is now read-only and reports a missing
vault rather than creating one. The global-state block moved below the
mode dispatch so it runs only on a default or --link run; the mutating
modes still need the vault repo, so they sit between the scaffold and it.
--adopt validates its slug before using it. It is interpolated into
"$vault/projects/$slug" and handed to `git mv`, so --adopt=../../../etc
walked straight out of the vault. Only the charset the slug formula emits
is accepted, with no slash and no leading dot.
--remote captures the git exit status explicitly rather than chaining an
`or` off the block terminator. That construct does work in fish, but it
reads as the silent-false-success shape that a hook-rejected commit once
produced here, and it stops working the moment the `else` goes away.
Also pins the dangling-global-symlink case the suite never covered: for a
broken ~/.claude/memory link both -d and -e are false, so the -L disjunct
in the global-memory guard is the only thing that notices it. That is the
state a buggy earlier run left on a real machine; the test asserts it is
detected, repinned into the vault, and exits 0.
The global block runs before the per-project link and the commit, but its
mkdir and link failures returned 1 outright. Global memory is optional and
frequently absent, so a stray file or a permission problem at
~/.claude/memory would abort the per-project memory backup and its commit
for every project, on every agent launch -- a fault in the secondary
feature killing the primary one.
Both failures now warn to stderr and continue, matching the treatment the
agy copy already had; the whole global block is best-effort by design.
Continuing is safe because _agents_repo_ensure_symlink validates and
refuses before mutating anything. $changed is set only when the link
actually succeeded, and nothing is recorded that would make a later run
believe the global memory is linked when it is not.
The live-side test widens from -d to -e so a stray regular file where the
global memory directory belongs is reported on every run instead of being
silently skipped and mistaken for the absent-by-default case.
Also documents that the agy knowledge copy is merge-only: a fact deleted
upstream persists in the vault and a restore brings it back. Whether the
vault should mirror deletions is a retention decision for the repo owner;
the gap is worth stating either way.
agy partitions by conversation UUID rather than by workspace, so it has no
per-project slice and is tracked globally. Its knowledge store is copied
rather than symlinked because it sits beside SQLite databases with WAL
sidecars. Claude's global memory directory is symlinked into the vault the
same way per-project memory is, including the emergent-restore direction.
Paths are allowlisted so credentials, transcripts, and session state cannot
be swept in.
Two variables keep the tests off the real home: the new
__fish_agent_vault_claude_home overrides ~/.claude (whose memory/
subdirectory is the global one), distinct from the existing
__fish_agent_vault_claude_root, which overrides ~/.claude/projects.
Without it a test run on a machine that has a real global memory directory
would move it into a mktemp vault and leave a dangling symlink behind. The
suite now points every run at a throwaway home by default and asserts the
real paths are untouched.
cp cannot report whether anything actually differed, so the copy is only
counted as a change when it leaves the vault's global/agy/ subtree dirty.
Marking it changed unconditionally would print a --quiet summary line on
every agent launch and make the flag meaningless.
The slug-migration fallback (used when there is no live symlink to read
the previous slug from) recomputed the local-* candidate by lowercasing
the basename only, while _agents_repo_slug sanitizes it. The two formulas
had drifted, so the fallback silently found nothing for any project
directory whose basename needed sanitizing.
Extract the formula into a single private helper,
_agents_repo_local_slug, and have both _agents_repo_slug's no-remote
branch and agents-vault's migration fallback call it, so there is one
place left to drift.
Also drop two dead lines the review flagged: an unused local,
and an unreachable mkdir -p (path dirname ...) — slugs never contain a
path separator, so dirname always resolves to a directory that already
exists by that point.
Widen migration test coverage: the current-entry-present-but-empty case,
a remote URL rewrite, a remote removal, and a dirty-basename fallback
test that fails without the sanitization fix and passes with it.
Adding a remote to a previously remote-less project changes its slug. Left
unhandled, the link step repinned the live memory symlink to a fresh empty
entry and orphaned the real memory.
The previous slug is read from the live symlink target rather than guessed,
which covers a remote being added, rewritten, or removed. When both the old
and new entries hold content the migration is ambiguous, so nothing moves
and the user is directed to --adopt.
config-help resolves keywords through this hand-maintained index; the
new "Agent Memory Vault" section in 07-customization.md had no entries
here yet. Adds agent-vault, __fish_agent_vault_dir, and
__fish_agent_vault_autopush, following the __fish_scrollback_history_dir
precedent (bare variable names as keys).
The guard around the symlink step only linked when the live Claude
project directory already existed, which is exactly backwards for the
clone-onto-a-new-machine restore case: a freshly cloned vault entry
would be silently left unlinked and a starting agent would write fresh,
history-less memory instead. _agents_repo_ensure_symlink already makes
its own parent directories and is idempotent, so nothing depended on
the guard; it is removed and the link is now attempted unconditionally.
Also stop swallowing a refused or failed link as success: the helper's
exit status is now checked, and agents-vault reports its own error and
exits 1 instead of silently continuing with no link in place.
Smaller fixes from the same review pass:
- check the exit status of _agents_repo_install_tools and the
core.hooksPath git config write, instead of discarding both
- give the vmem mkdir failure a stderr message like every other fatal
in the function
- guard hostname with type -q and add it to DEPENDENCIES
- .version creation now sets changed, so --link (which skips the
commit step) reports it in --quiet mode
- reword --link's help/doc text: it still scaffolds the vault and
links memory, it only skips the final commit
- drop the unused c_dim color variable
- move the __fish_agent_vault_dir / __fish_agent_vault_autopush
documentation below Opinionated Components so its NOTE: callout
(now flush-left so it actually renders as a Starlight Aside, per
review) doesn't become the first Note aside in the page and steal
the existing test's assertions about the original 4-bullet one
Adds two tests: pre-seeded vault entry with no live directory at all
(the restore path the guard was breaking), and a forced link failure
asserting agents-vault now exits 1 instead of 0.
Creates the vault repo on demand, reusing the AGENTS version bumper and
hook shims, then links the current project's live memory directory into
its slug-keyed entry and commits.
Because the live directory becomes a symlink into the vault, backup and
restore are the same operation: a cloned vault relinks itself on the next
run in each project, with no manifest and no batch restore step.
Also fixes _agents_repo_install_tools' progress messages, which hardcoded
the literal "AGENTS/.agents-tools/" even for callers writing elsewhere:
they now name repo_dir's own basename, so agents-vault reports its own
directory instead of a false AGENTS/ path.
Renames _agents_init_install_tools to _agents_repo_install_tools now that
the vault shares it, collapses the two duplicated root-symlink blocks into
one loop, and routes the auto-commit through _agents_repo_sync so a failed
rebase can no longer be committed as conflict markers.
_agents_repo_sync fell off the end of its final if-block when git commit
failed, e.g. a pre-commit or commit-msg hook rejecting it (this repo runs
ggshield and Git-LFS hooks). fish's if construct sets status 0 when the
condition is false and there is no else branch, so a rejected commit was
being reported as success rather than as the documented "not a git
repository" exit 1 it was assumed to fall through to.
Add an explicit else branch that emits a stderr diagnostic and returns 1,
and widen the EXIT STATUS/DESCRIPTION docs to cover this path under the
existing code 1 rather than adding a fourth code, since later tasks
already consume the 0/1/2 contract.
agents-init currently swallows a failed rebase and then stages and commits
whatever is in the tree, which records conflict markers under a routine
message. No AGENTS repo has a remote today so the pull never runs, but the
vault gives these repos remotes and arms it.
The shared helper aborts the rebase, commits nothing, and returns 2. It
also redirects git's own stdout during the pull/abort: git prints
"CONFLICT (content): ..." to stdout, not stderr, so without this the
message would leak into the helper's own stdout instead of staying
diagnostic-only.
The conflict fixture commits "ours" locally before diverging, since an
uncommitted worktree change has nothing for --autostash's rebase step to
replay -- it fast-forwards cleanly and only the stash pop would conflict.
Enforces the rails the vault depends on: only directories are linked
(agent editing tools refuse to write through a symlinked file), a missing
target is refused rather than turned into a dangling link, and adopting a
populated live directory copies without clobbering.
Also fixes slug sanitization in _agents_repo_slug to apply the same
[^a-z0-9._-] → - mapping to the fallback (no-remote) branch, ensuring
local project slugs are filesystem-safe and won't leak special chars like
spaces or exclamation marks.
Keys a project by its remote rather than its path so the key survives a
machine change or a directory rename. Falls back to a path-derived
local-* key when no remote exists.
Adds a hermetic test harness that builds throwaway repos under mktemp.
The mirror section told contributors a label added on Gitea "must be
created on the mirror too -- no automation does it for you." That's no
longer true, and a stale instruction to do something by hand is worse than
none, since it invites a manual edit that the next scheduled sync would
overwrite anyway.
Describe what the sync actually does: what it creates, updates, and prunes,
that an in-use label is never deleted, the --dry-run and --self-test flags,
and the rename caveat that follows from matching labels by name. Also
record the GH_MIRROR_TOKEN secret the workflow needs and its exact scopes,
since that's the one part of this that can't be automated.
The exclusive-labels caveat below it is unaffected and stays as written.
Labels don't travel with a mirror push -- mirroring copies files, not
repository settings -- but they matter on the GitHub side anyway, because
GitHub reads the same .github/ISSUE_TEMPLATE/ files and silently drops a
labels: entry naming a label it doesn't have. Until now the only thing
keeping the two sets aligned was remembering to do it by hand, which is
exactly the kind of thing that gets forgotten and then fails invisibly.
Add scripts/sync-labels.py and a workflow that runs it daily, on any change
to the script itself, and on manual dispatch. Gitea stays the source of
truth: labels are managed there and GitHub is made to match.
- **Creates and updates** anything missing or drifted. Colors and
descriptions are normalized before comparison -- Gitea returns colors
bare, GitHub sometimes with a leading '#', and a description may be null
on one side and "" on the other -- so a steady state is a true no-op
rather than a rewrite of all 33 labels every run.
- **Deletes only unused extras.** An extra label on the mirror is removed
only when no issue or PR there carries it; one in use is reported with
its count and left alone. An unattended scheduled job must not be able to
strip a label off somebody's issue.
- **Refuses to run on an empty source**, since treating that as truth would
propose deleting every label on the mirror.
- **--dry-run** prints the plan and changes nothing; **--self-test** checks
the diff logic offline against fixtures, and gates the sync step in CI so
a broken diff can't mutate anything.
Stdlib only, so the CI step installs nothing beyond python3. The job is
gated with the same github.server_url check ci.yml uses -- without it the
mirror would queue this forever against a runner that only exists on Gitea.
Labels are matched by name, so a rename reads as delete-plus-create; the
new name is created and the old is pruned only if unused. The two forges
share no stable label ID, so a rename can't be tracked across them.
The templates were written against Gitea's schema alone, but the mirror
serves the same .github/ISSUE_TEMPLATE/ files to GitHub, where two of them
would have silently failed:
- **config.yaml -> config.yml.** Gitea accepts either spelling
(modules/structs/issue.go: `base == "config.yaml" || base == "config.yml"`),
GitHub only recognizes config.yml. Under the old name the mirror's
template chooser would have shown neither the contact links nor the
blank-issue setting.
- **bug.yaml `about:` -> bug.yml `description:`.** GitHub requires
`description` on a YAML issue form; Gitea's IssueTemplate.About carries
the comment "Using description in a template file is compatible" and
falls back to it at modules/issue/template/unmarshal.go:126. So
`description` is the one spelling both forges accept. The markdown
templates keep `about:`, which is correct for their format on both.
Both files now carry a comment explaining the constraint, so neither gets
"tidied" back into a broken state.
Also add a contact link pointing at the canonical Gitea tracker, so someone
arriving from the mirror is steered to the right place before they file,
and document mirror parity in CONTRIBUTING.md: the labels must be created
on both sides by hand, since mirroring copies files rather than repository
settings and GitHub drops a labels: entry naming a label it doesn't have.
Note too that GitHub has no exclusive labels, so the one-of rule on
Priority/, Reviewed/, and Status/ holds only by convention there.
The repo carried Gitea's stock label set with no written rule for applying
it, so labels were effectively unused. Establish the standard: every issue
and PR carries exactly one Kind/ and at least one Area/, and document what
each group means, which are exclusive, and who applies them.
Extend the taxonomy to make that workable:
- **Kind/Refactor, Kind/Chore, Kind/Performance** — the stock Kind/ group
couldn't describe a refactor, a chore, or a perf change, which is much of
this repo's history. The group now maps one-to-one onto the Conventional
Commits types already in use, so a PR's label and title agree.
- **A new non-exclusive Area/ group** over the ten subsystems (functions,
completions, config, docs, tests, CI, integrations, prompt & theme,
components, scripts). This is what makes the tracker searchable by
subsystem; Kind/ alone can't answer "what's outstanding in the docs
pipeline?".
- **good first issue and help wanted**, deliberately unscoped rather than
under Status/. Status/ is exclusive, and an issue is often both blocked
and open for someone to pick up; keeping these outside the group lets
them coexist with a real status.
Priority/, Reviewed/, and Status/ are documented as exclusive and
maintainer-applied, with the rules that keep them meaningful — no
Priority/ label means ordinary priority, Reviewed/Confirmed means actually
reproduced, and a Status/ label is removed once it stops being true.
Also add labeling to the standing PR rules in Branching & Pull Requests,
so it's visible where the rest of the PR checklist lives.
Issues had no template at all, so a report arrived in whatever shape the
reporter chose — most often without a fish version, a reproduction, or the
full error text, which is what actually stalls a bug.
Add three templates under .github/ISSUE_TEMPLATE/, beside the PR template
so the GitHub mirror offers the same set:
- **bug.yaml** — a Gitea issue form rather than markdown. Version, OS,
area, reproduction, expected and actual behavior are required fields, so
an unactionable report can't be submitted in the first place. The Area
dropdown exists because contributors without push access can't set an
Area/ label themselves.
- **feature.md** and **docs.md** — comment-guided markdown in the same
house style as PULL_REQUEST_TEMPLATE.md, since what they ask for is
open-ended prose. feature.md carries `## Acceptance criteria`, the
issue-side counterpart to a PR's `## Verification`. docs.md insists on
the docs/manual/** source rather than the generated page, which the next
CI run would overwrite.
- **config.yaml** — keeps blank issues enabled for what the three don't
cover, and links the contributing guide and the customization docs.
Each template pre-applies its Kind/ label. Document the set, the plain-
description title convention (an issue states a problem; the conventional
subject belongs on the PR that closes it), and the triage split in a new
CONTRIBUTING.md § Issues.
Gitea treats a `WIP:` title prefix as a draft marker and refuses to merge
while it's present. Verified empirically against this instance by
temporarily retitling PR #123: the API flipped `draft` to true and
`mergeable` to false, and both reverted when the prefix was removed.
The distinction from the verification merge gate is the point, so it's
stated explicitly in both files: `WIP:` means more code changes are
coming, NOT "finished but unverified". A complete branch with unticked
`## Verification` boxes is an ordinary PR — that case is already covered
by the merge gate and does not want the prefix. The two signals are
independent and can occur in any combination.
Documented as a merge rule under Branching & Pull Requests, next to the
verification gate, and in the template's title guidance where an author
picking a title will actually encounter it.
The old name implied the list was only for by-hand steps, which sat badly
with the fact that most entries in practice are programmatic — the test
suite, `fish_indent`, `docs/verify-manual.py`, CI. Dropping "Manual" lets
one list carry both kinds, and gives the checkbox state real meaning:
- **Checked** — verified, whether it ran programmatically or the author
performed it by hand.
- **Unchecked** — an outstanding manual check the reviewer still has to
perform. Left visible rather than dropped.
- **The list is the merge gate.** A PR isn't merged until every box is
checked; added as an explicit rule under Branching & Pull Requests
alongside the other merge rules.
Because unchecked boxes now block the merge, the guidance also states that
only resolvable checks belong here — one nobody can run would block the PR
indefinitely. Genuinely unverifiable caveats and assumptions go in
`## Notes`, which already existed in the optional-sections list.
Incidentally realigns with PRs 121-122, which had already shortened the
heading to `## Verification`; that part of the recent drift was right.
The repo has no issue-tracking history yet — a grep across all 122 prior
PRs found zero `Closes #N` references — so this is forward-looking rather
than derived from precedent. Establishing it now means the first PR that
does close an issue has a rule to follow instead of inventing one.
- Placement is a trailing line at the end of `## Summary`, not the bottom
of the body, because `## Manual Verification` is always last.
- `Fixes`/`Resolves` noted as equivalent; `Refs #N` for a related issue
that should stay open.
- Guidance calls out that the keyword must be repeated per issue, since a
bare `#43` after a comma links without closing.
- Lives in the template's HTML comments, so a PR with no associated issue
leaves no stray `Closes #` behind.
Also drops an unverified claim that the GitHub mirror pre-loads the
template; PRs are opened on Gitea, and the mirror is push-only.
Codifies the PR body format the repo has converged on across its first
122 pull requests, which until now lived only as an implicit pattern
agents and contributors had to reverse-engineer from prior PRs.
- **`.github/PULL_REQUEST_TEMPLATE.md`** — `## Summary`, optional `##`
sections, then `## Manual Verification` as a checkbox list. Gitea falls
back to `.github/` when `.gitea/` is absent, and the GitHub mirror reads
the same path, so one file covers both.
- **`CONTRIBUTING.md`** — new "Pull request descriptions" subsection under
Branching & Pull Requests, documenting the same three-part structure.
Section names were chosen by frequency across all 122 PRs: `## Summary`
(86) and `## Manual Verification` (63) are the clear majority. The later
`## Manual Verification Checklist` (12, PRs 90-113), `## Test plan` (11)
and `## Verification` (3, PRs 121-122) variants are drift away from that
baseline, not a newer standard, so the template restores the dominant
form. Checkboxes ship unchecked but are meant to be checked before the
PR opens, matching the 585-to-12 ratio of `[x]` to `[ ]` in merged PRs.
The heading scan matches a keyword contained in a heading, so a plural
could never reach a singular heading: `customization` found "7.
CUSTOMIZATION" and `customizations` found nothing at all. The index
lookup was exact-match, so it could not cover the gap either without an
alias per word.
Both now try the keyword as typed first, then its singular/plural
forms. Each variant is tried against every heading before the next one
is considered, so a loose plural cannot beat an exact hit further down
the document.
config-help pipes the manual through bat, which dims the backticks and
leaves the span content the same colour as the prose around it -- so a
delimiter carried no information and every span cost the reader two
literal characters. 1236 of them across the document.
Each span is now rendered bold and the delimiters dropped, in every
branch of the viewer chain:
- ov + bat, and ov alone, style the spans (bat's output wraps each
backtick in its own SGR sequence; raw Markdown is matched directly)
- bat alone flattens them on the way in instead, because bat escapes
any SGR sequence handed to it as input
- less and cat style them, less gaining -R to render the result
- man -l needs nothing; pandoc consumed the backticks at build time
Both substitutions are line-preserving, so the tail-slice that opens
the pager on a requested section still lands on it.
Two shapes reach a reader as literal punctuation rather than markup:
A backtick inside a four-space block. The block is verbatim in every
renderer -- pandoc sets it monospace, prettify() fences it for the site
-- so the backtick is a character on the page. Twelve such lines are
cleaned; each sat at the start of its column, so the alignment of the
C5 capture table and the component summaries is unchanged.
A span split over a line break. Markdown pairs it happily, but
config-help pairs backticks one line at a time, so `fish-deps\nupdate`
showed both halves. The sentence is reflowed.
Both are now enforced, the first by test_concat_section_five_stays_
verbatim and the second by test_concat_code_spans_never_straddle_a_line.
A token was typeset by whichever pipeline happened to render it: the
site marked tmux and local.fish through codespans, while the man page
and config-help marked only what the SSOT had backticked by hand. Run
the same pass in build_concat() so prose is marked identically wherever
it is rendered (549 -> 662 spans in the concat).
codespans now treats a four-space block as code. The site never meets
one -- prettify() has already turned it into a fence by then -- but the
concat keeps the indented form pandoc wants, and its contents are
verbatim: without this the table of contents alone would come out with
ov, bat, less and cat wrapped inside a code block. Section 5 is
unaffected for the same reason; its entries arrive as indented blocks
that pandoc already sets in a monospace font.
test_codespans_is_site_only asserted the opposite guarantee and was
passing only because its example, -r/--resume, sits inside one of those
newly-protected blocks. It is replaced by tests for what is now true:
indented blocks stay verbatim, prose spans reach the concat, and
section 5 carries no backticks.
The man page is left for CI to regenerate; pandoc is not needed to
build the concat.
CONTRIBUTING states doc-headers are written as plain text -- the header
is read as-is by config-help, by funcsave, and by anyone opening the
file, and docs/codespans.py adds the site's inline code spans at render
time. 22 files had drifted from that, carrying 41 hand-written spans
that reached config-help and the man page as literal backtick
characters inside an otherwise verbatim block.
The one span whose content ended in a space is requoted rather than
dropped, so "read> " keeps reading as a prompt string.
Cover the ~50 previously-unindexed headings (function reference entries
like play-media, steam-dl, bd-pull, cffetch/ffetch, config-toggle, dops,
rand_string, fish_prompt/fish_mode_prompt/fish_right_prompt, lD, mv, p,
fast/fast-cli, gip4/gip6, fzf-update, sponge_filter_secrets,
fzf_configure_bindings, ld; the C1-C6 component sub-category headings
filesystem/network/monitor/shell-tools/dev-tools,
plugin-management/pkg-wrappers/venv/telemetry/sync,
key-bindings/environment/prompt, terminal-abbrs/window-mgmt/
notifications/history-logs/pkg-upgrade, terminal-capture/
multiplexer-capture/pkg-logs, first-run/greeting-message; plus
Config Variables' "Other", the Integrations "Scrollback History", the
dependency catalog's Optional/Terminal Emulators tiers, the C0/always
override note, and "Reading the source directly").
Expanded aliases for existing headings: play-media also reachable as
video/audio/mpv/vlc, steam-dl as steam, cffetch/ffetch share fetch/
fastfetch/neofetch/sysinfo, rand_string as random/password/randomword,
gip4/gip6 as ipv4/ipv6, dops as docker-ps, ld as lazydocker, bd-pull as
beads, fast-cli as speedtest, sponge_filter_secrets as secret-filter.
Fixed two pre-existing duplicate normalized keys (key-bindings/
keybindings and man-page/manpage both collapsed to a single entry) and
repointed config-toggle from the generic config-settings heading to its
own dedicated "### config-toggle" heading now that one exists.
Judgment calls on ambiguous/repeated heading text (left unindexed, or
indexed via a distinguishing keyword instead of the literal name):
- "### Sub-categories" repeats 7 times (once per opinionated-components
overview and once per C1-C6 page); an index entry pointing at that
text always resolves to the first occurrence, so none of the 7 are
indexed under that name. Each parent (C1-C6, and the Minimal Mode
section) already has its own keyword.
- "### search" appears twice: the pkg search subcommand (kept, existing
`search` key) and the C1 command-shadow sub-category. The C1 one
isn't separately reachable for the same reason.
- "## Integrations" appears twice (Section 1 integrations writeup and
the Dependency Catalog's Integrations tier table); only the first
(already indexed) is reachable.
- Where a C1-C6 sub-category heading text collides after normalization
with an existing higher-value keyword (key-bindings, network,
notifications, venv), the new entry uses a distinguishing keyword
(bindings-toggle, network-shadow, notifications-toggle, venv-toggle)
instead of contesting the existing one.
- `prompt` now points at the fuller "## Prompt and Theme" section
rather than the one-line C3 sub-category blurb; the latter is
reachable via `prompt-toggle`.
The test/build-docs jobs target a self-hosted runner (racknerd-mini)
that only exists on the Gitea instance. When GitHub re-runs this same
workflow on the mirror, those jobs sit queued forever with no matching
runner, so the commit never gets a completed status.
Gate both jobs to skip when github.server_url is github.com, and add a
trivial github-mirror job (runs-on: ubuntu-latest, which GitHub does
provide) that only runs on the mirror, so the check completes instead
of hanging.
Function doc-headers are authored as plain text -- `config-help`,
`funcsave` and anyone opening the `.fish` file read them as-is -- so they
carry no backticks. The site inherited that and rendered `-a/--all` and
`__fish_config_op_aliases` as ordinary prose.
docs/codespans.py adds the spans at render time, as the last step of
prettify(), so only the site sees them; build_concat() (man page,
config-help) is byte-for-byte unchanged.
Recognised shapes: flags and flag pairs, `$vars`, SCREAMING_SNAKE env
vars, snake_case identifiers, paths and filenames, key chords, command
shadow chains (`ls->eza`), runs of tool names, whole command lines in a
table column of command lines, and known command names -- drawn from the
`_fdc_*` catalog in functions/_fish_deps_catalog.fish, the functions/
listing, and a standard-command list, minus the names that also read as
English.
Fenced blocks, existing code spans, headings, link targets, URLs,
component markup and <FileTree> bodies are passed through untouched, and
every rule bails out rather than guess.
Verified every claim in the Prompt and Theme section against the actual
source and corrected several inaccuracies: the Starship wrapper's missing
C3 gate, the fallback prompt's vi-mode states and segment order, the FZF
theme's real location (conf.d/theme.fish, not integrations/fzf.fish) and
color set, and the Catppuccin theme-switch example using the wrong
fish_config subcommand (choose, not save). The right-prompt Docker-context
example was rewritten to show that it's independent of exit status.
While reformatting, found that any block build-manual.py couldn't classify
as shell/table/tree fell back to plain 4-space markdown indentation, which
silently renders as squashed, unreadable paragraph text on any page that
also contains an <Aside> or <FileTree> — MDX has no indented-code-block
syntax, unlike plain Markdown. This affected 07-customization.mdx plus four
other pages. Fixed the fallback to emit a fenced ```text block instead,
since fences work in both MDX and plain Markdown; this also gives every
affected block Starlight's normal code-block styling instead of a bare grey
slab.
docs/fish-config.md is intentionally left stale here — CI regenerates and
auto-commits it from docs/manual/** on push to main.
List every file in functions/ and completions/ inline under the home
page's file tree instead of a one-line summary. The listing is read
live off disk during --site generation, so it never needs manual
upkeep, and only affects the Starlight build — the plain-text
manual/man page (--concat) still renders the compact summary.
rootiest
Gitea Actions