Follow-up cleanup on the publish-failures change.
Error model:
- Collapse `Error::{Engine, Publish, Handler}` into one `Error::Stage` with an
`ErrorStage` discriminator. The three shared a field shape and had to be
edited together in four match groups; nine near-identical constructors
become two private helpers.
- Add `Error::failure_reason()`, replacing the same error -> FailureReason
mapping written out in four places.
- Publish errors are now terminal. Publish runs once, after execution, so no
caller could ever act on the retryable classification.
Publish phase:
- Fix: a branch that was pushed is now still reported when pull request
creation fails afterwards. `PublishOutcome` records what happened and
carries the error separately, instead of hiding both behind a `Result`.
- Drop `PublishOutcome::NoChanges`, which no consumer distinguished from
`Published { pr_url: None }`.
- Move publish onto `Concluded` as methods and replace three near-identical
precondition guards with one `publish_target()`.
Pull requests:
- `maybe_open_pull_request` -> `open_pull_request` returning the record
directly. Both callers already reject empty diffs, so the `Ok(None)` path
was unreachable.
- Drop `CreatedPullRequest.head_sha`, which echoed back its own input.
GitHub client:
- Delete `branch_exists`, which had no callers and duplicated
`branch_head_sha`. Give `branch_head_sha` the `_with_client` split every
sibling has and port the tests to `MockHttpClient`.
- Collapse the copy-pasted credential match in `resolve_clone_credentials`.
Events:
- `PullRequestCreated.head_sha` is `Option<String>` instead of using an empty
string to mean absent.
- Centralize the run-branch refspec in `lifecycle::push_run_branch`, so
`git.push` reports a branch name from both emitters as documented.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Fabro advertised Bash while its three backends implemented three
different contracts: Daytona evaluated commands through `sh`, and
Docker's streaming, stdio, and setup paths used a login shell. Bash-only
syntax silently misbehaved depending on provider and code path, and
login profiles could change PATH and command behavior per image.
Make `bash -c` the enforced interpreter for every command string the
Unix sandbox API accepts, on every production backend and through both
buffered and streaming execution. This selects the interpreter only —
no `errexit`, no `pipefail`, no login mode — so `false | true` still
succeeds and a workflow that wants other semantics writes them into its
own command.
Local resolves `bash` through the worker's PATH (NixOS has no
/bin/bash) and reuses that one executable across all three command
paths. Docker and Daytona require /bin/bash with no `sh` fallback.
Fresh initialization and resume/start now verify Bash through a shared
marker-validating probe before reporting the sandbox usable, so a
missing or non-Bash interpreter fails at the lifecycle boundary with
provider-specific remediation instead of on the first command. The
probe also rejects Bash in POSIX mode, which an image whose `bash` is
really `sh` would otherwise pass.
Sandbox MCP scripts and the detached launch wrapper move under the same
contract; host-side stdio MCP scripts, hooks, and interactive terminals
are separate executors and keep their existing `sh` behavior.
The `shell` tool's name and JSON schema are unchanged across providers;
only its prose now identifies `command` as Bash source.
BREAKING CHANGE: sandbox commands no longer load login-shell profiles,
so environment set in /etc/profile.d/*.sh, ~/.bash_profile, or
nvm/rbenv/sdkman initializers is gone. Move those exports into the
Dockerfile's ENV or the Daytona snapshot image.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The previous commit introduced render_prompt and splice_optional_section, a
second templating mechanism in a workspace that already standardizes on
MiniJinja behind fabro-template. Drop both and render the profile prompts the
same way fabro-workflow and fabro-manifest render theirs.
Expressing the conditionals as {% if %} lets every profile collapse to a single
template, since the optional blocks no longer need to be separate files spliced
in from Rust:
before: 6 files + 2 splice helpers, prompt prose split across .md and .rs
after: 3 files, one per profile, all prose in the template
Rust now passes only facts -- provider name, which file-edit tool is active,
and whether web search and subagents are available. Values land under `vars`,
so templates read {{ vars.env_block }}. Booleans are passed as "true"/"false"
and compared explicitly via the bool_var helper, because the shared
TemplateContext types vars as strings and a bare {% if %} on the string
"false" would be truthy.
Also converts fabro-server's Ask Fabro prompt, which is assembled at runtime.
Its tool guidance now arrives as a template variable instead of being
interpolated into the template text. That guidance carries tool names and
descriptions that can originate from MCP servers, and MiniJinja does not
re-render substituted values, so a tool description containing {{ ... }} stays
inert rather than being evaluated.
Output is unchanged. Verified by diffing all ten prompt variants against the
same unmodified origin/main worktree used for the previous commit.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Return 400 (not 500) for WorkflowError::ModelReference from run
creation, matching ModelSelection: an ambiguous model/provider token
is user input, not a server fault.
- Gate fabro-workflow's test_support module behind
cfg(any(test, feature = "test-support")) so the feature actually
controls exposure, per the repo's test-support boundary guidance.
Add the self dev-dependency so tests/it keeps compiling, and gate
the pipeline helpers that only test_support consumed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Consolidate duplicated logic from the SQLite runs read model review:
- Derive the status sort CASE and board-column filter from a new
RunStatusKind::board_rank(), replacing three hand-maintained copies
of the status/column mapping; add a test upserting every status
variant so the migration CHECK can't silently drift
- Share RunSize bucket thresholds between from_total_usd_micros and
the generated size-sort CASE via RunSize::BUCKET_MAX_USD_MICROS
- Resolve run selectors from a lean identity query instead of
decoding every stored summary per request
- Delete the RunsSortKey/RunsSortDirection adapter enums; the store
sort enums now carry the wire serde names
- Consolidate the workflow display-name fallback chain into
WorkflowRef::display_name() (store, CLI, run lookup)
- Share pagination clamping and the paginated list envelope across
handlers
- Reconcile now skips rows whose source seq is unchanged and
batch-deletes stale rows; drop the two indexes no query can use
- Hold the summary store OnceLock cell in RunDatabaseInner instead of
a snapshot so late attachment reaches already-open writers
- Misc: expect() on COUNT(*) sign, %err logging, shared wall-time
helper, shared SQLite test fixture, dead billing fallback removed
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Delete the dead test-only Vault-based env-secrets migration and point
the startup migration tests at the production migrate_to_store path
over a real SQLite-backed SecretStore
- Extract shared legacy-import helpers (timestamped backup rename,
is_toml_file) into fabro_db::legacy and parse_rfc3339_utc into
fabro-db, replacing four per-crate copies
- Take one secrets snapshot in migrate_to_store instead of per-name
queries
- Share one bind order between the MCP store INSERT and UPDATE
statements
- Return SecretEntry directly from entry_from_row
- Unify the environment/MCP store blocking loaders into a generic
load_store_blocking helper
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Review pass over the secrets-to-SQLite migration:
- Add SecretStore::open() consolidating the connect/migrate/import-legacy
sequence repeated at five call sites; fabro-agent and fabro-cli drop
their fabro-db dependency
- Restore process-env LLM credential lookup in the standalone CLI/agent
sources via SqlVaultCredentialSource::new (regression: vault_only
dropped the env fallback that VaultCredentialSource::new provided)
- Fix five install tests that still asserted against the legacy
secrets.json, which the importer renames to .bak
- Make AppStateConfig.preloaded_vault required, deleting the fallback
that re-read the already-renamed legacy file; drop the now-unused
vault_path field and demote load_startup_vault to test-only
- Skip the snapshot clones and CAS retry in resolve() when the vault
holds no OAuth secrets (per-request hot path)
- Remove dead persist_with_secret_store, the VaultSecretWrite alias,
the secret_type_string one-liner (now SecretType::as_str), the
impossible RowCountOverflow error, and duplicated row parsing
- Run check_crypto concurrently with the other diagnostics checks
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A binary downgrade after new SQLite migrations have been applied fails
sqlx's startup validation ("migration was previously applied but is
missing in the resolved migrations") and previously left the operator
with no rollback artifact: the shared database had no backup, so
recovering meant hand-editing _sqlx_migrations and dropping tables.
Database::migrate now writes a consistent single-file snapshot to
<db>.pre-migration.bak (via VACUUM INTO, mode 0600) before applying any
migration the database has not seen. Rollback is: stop the server,
replace the database file with the snapshot, delete -wal/-shm siblings,
start the previous binary. Fresh databases and no-op migrates skip the
snapshot, so the file always preserves the state from immediately before
the most recent schema change. A snapshot failure fails the migration:
no rollback artifact, no schema change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
## Problem
Loading the web UI from a remote server took **~11 seconds to first
render on every refresh**. A HAR capture against a remote deployment
showed the page downloading **13.5 MB of JavaScript across 356 files,
uncompressed, on every single page load** — even though the assets are
content-hashed and served with `Cache-Control: immutable`.
Four compounding causes:
1. **`Pragma: no-cache` defeated the browser cache.** The
security-headers middleware stamped `Pragma: no-cache` onto every
response, including hashed assets that set a year-long immutable
`Cache-Control`. Browsers treat a response `Pragma: no-cache` as
`Cache-Control: no-cache` and check it *before* `max-age` (Chromium
zeroes freshness on it), and since assets carried no validators,
"revalidate" degraded into a full re-download. Empirically visible in
the HAR: Google-Fonts woff2s served from cache (`transfer = 0`) during
the same page load where all 356 of our assets re-downloaded in full.
2. **No response compression.** The server had no compression layer;
13.5 MB of JS compresses to ~2.5 MB with brotli.
3. **The HTML force-loaded every chunk.** `writeIndexHtml` emitted a
`<script type="module">` tag for all 356 outputs. Only 2.9 MB is
statically reachable from the entry; the other ~10.7 MB is
dynamic-import-only code (syntax grammars, Graphviz WASM, xterm, diff
file tree) that was being downloaded eagerly at high priority.
4. **The immutable heuristic over-matched.** Any dash in a filename
counted as a content hash, so stable-named files
(`pierre-diffs-worker/worker-portable.js`, `apple-touch-icon.png`) would
be pinned in browser caches for a year across deploys once fix 1 made
immutable caching effective.
## Changes
- **`security_headers`**: apply the `no-store`/`Pragma: no-cache`
defaults only when the handler didn't set its own `Cache-Control`. API
responses keep the conservative defaults.
- **Compression**: `tower-http` `CompressionLayer` (brotli + gzip) on
both the main router and the install-mode router (install mode serves
the same SPA bundle through a separate router). Default predicate keeps
SSE (`text/event-stream`), gRPC, images, and tiny bodies
identity-encoded. Quality pinned to `Precise(4)` — tower-http's default
defers to the codec default, and brotli's default is quality 11 (seconds
of CPU per multi-megabyte asset).
- **Entry-only HTML**: `writeIndexHtml` emits script tags only for `kind
=== "entry-point"` outputs. The module graph pulls static imports (depth
1, so no waterfall); dynamic `import()` chunks load on demand.
- **Cache-control classifier + validators**: only files matching the
bundler's actual output shape (`assets/<stem>-<hash8>.js|css`, lowercase
base-36) get `immutable`. Everything else is `no-cache` **with a strong
ETag** and `If-None-Match` → `304` support, so index.html / app.css /
the pierre worker revalidate in one cheap conditional request instead of
a full re-download.
## Impact (measured on the built bundle)
| | Before | After |
|---|---|---|
| Cold load, ~1 MB/s link | 13.5 MB raw ≈ **11–14 s** | ~0.8 MB
compressed eager payload ≈ **~1 s** |
| Refresh | full re-download, same 11–14 s | served from cache + one 304
≈ **instant** |
| Eager JS on first render | 13.56 MB / 356 files | 2.88 MB raw (0.79 MB
gzip) / 6 files |
## Verification
- 959 fabro-server tests pass (incl. new coverage); fmt + clippy clean;
`bun run typecheck` passes (the 5 pre-existing bun test failures
reproduce identically on `main` — missing `@pierre/diffs/dist/worker`
fixture + flaky InstallApp timing tests).
- New integration tests pin compression through **both** serving shapes
that matter: regular routes and the SPA fallback service, each via tower
`oneshot` **and** over a real TCP connection through hyper (raw-socket
assertions, so no client auto-decompression can mask a regression).
- Live-verified against a debug server: hashed assets get `immutable` +
brotli and no `Pragma`; mutable assets get `no-cache` + ETag and answer
conditionals with `304`; API responses keep `no-store`.
- Headless Chrome boots the rebuilt SPA from the entry-only HTML and
fully renders the UI.
## Notes for reviewers
- The ETag is skipped for immutable assets deliberately — they never
revalidate, so hashing multi-MB bodies per request would be pure
overhead.
- Install mode previously had **no** compression and shares the same
bundle; it gets the same layer via a shared `compression_layer()`
helper.
- `bun test` has a pre-existing suite (`production build copies Pierre
worker assets`) that fails without `@pierre/diffs/dist/worker` present
locally; unrelated to this change.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
