[huyplb]

Three separate pieces of work, each its own commit. The first is the security fix this PR opened with; the other two are the 0.3.1 release flag and the new Utilities panel, folded in because they share the branch.

1. fix(sql): a name's line break escaping a generated comment

The script writes names into -- notes — the per-step header, review notes, manual-review blocks — and a comment ends at the first line break. A name carrying one closed its note and left the rest of itself behind as a statement the migration then ran:

-- CREATE VIEW: v
DROP TABLE victim; --
CREATE OR REPLACE VIEW "v DROP TABLE victim; --" AS
SELECT 1;

Quoting doesn't help: a line break can't be escaped inside a quoted identifier. Dialect-independent, and reachable the same way as the other identifier injections — a source schema the user doesn't control.

Names reach comments by three routes, so the fix has three layers:

Layer	Covers
flattenIdentifier in ident	every name the generator quotes (ident / bareName / qualify / columnIdent)
MigrationStep.objectName, flattened at all 4 assignment sites	the per-step header, emitted separately from the statements — the route the first reproduction took
oneLineComment, plus keepCommentCommented as a backstop	names ident never sees (index / constraint / trigger, type warnings), and notes a dialect hook builds from values the generator never quotes

That third layer matters: ClickHouse's ENGINE / ORDER BY note repeats the sorting key and engine string, SQLite's column note repeats the type. All three were exploitable, and the backstop covers hooks added later. It can't break DDL: it only adds -- to a line, and only inside a statement that already opens with --. Verified a no-op across 1056 statements from every dialect, same- and cross-dialect.

Swept rather than spot-checked: 99 shape × dialect combinations with the payload in every name field, 35 generateObjectDdl combinations, and line breaks in a sorting key, engine string, column type and column default. 10 of the 11 new tests fail on the unfixed generator; the eleventh asserts ordinary output is unchanged and passes either way.

Trade-off: a name that really does contain a line break is emitted with spaces, so that object needs a hand-written migration. Such a name can't be escaped in SQL and is pathological — the alternative is running DDL a source schema chose. The name still reaches the DDL flattened, so the object stays identifiable.

2. chore(web): Workflow off the rail for 0.3.1

One flag, NAV_FLAGS.workflow, takes Workflow off the activity rail and out of the command palette. The view's code and /api/workflow are untouched and the engine still answers, so flipping it back to true is the whole of putting it back.

A view is reached only through the rail or the palette — there are no URLs for views — but activeView is persisted, so a session that was on Workflow before the flag went off reopened on it. redirectFor now also sends anyone home from a view whose nav item a flag hides, following filterNav's own rule. The E2E specs that drive the view skip on a mirrored WORKFLOW_ON_RAIL constant and run again unchanged with the flag.

3. feat(utilities): which lines name the column being renamed

"View v_open_orders may use it" says there is work without saying where. The plan now carries, for each dependent view, function, procedure and trigger, the lines that name the column — with the match offsets — and the panel shows each line as written, highlighted, and (when renaming) as it will have to read. Read-only on purpose: Fox renames the column, it does not rewrite somebody else's routine.

One matcher in columnReferenceLines produces both the highlight offsets and the rewritten line, so the two cannot disagree. The existing name lists are untouched — the warnings and summary still read them — and one shared filter feeds both, so they cannot drift.

Two defects found reviewing it afterwards, fixed in the last commit:

• The "after" line is presented as what the SQL needs to say, so it must parse. Renaming status to order status put the bare name in while the plan's own ALTER quoted it. The new name is now quoted for the dialect when it needs it; where the definition already quotes the old name those quotes stay and only the word inside changes, so nothing is quoted twice. A plain name is inserted untouched, so every ordinary rename reads exactly as before.
• A generated view can name a column on hundreds of lines, each a row React built. The list stops at 50 and says how many are left; the count above still reports them all, and the whole definition is one click away.

Checks

• npx vitest run — 6948 passed, 1 expected fail, 145 skipped
• npx eslint . · npm run lint:security · npm run typecheck — all exit 0
• Each of the 5 review-fix tests verified to fail without its fix

Not in this PR

table.engine is still interpolated raw into the ClickHouse ENGINE clause, so a hostile engine string yields two statements. Engine names come from a fixed set, but engine parameters (ReplicatedMergeTree('/path','{replica}')) are chosen at CREATE TABLE time, so validating it needs care not to break parameterised engines — left for its own decision. The comment backstop here does cover the engine string where it appears inside a -- review: note, just not in the executable clause.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_019ehcTb7vYMCawV1ajxwfxB

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: a84ab433-f497-4934-a210-6f4cb8bfb544)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 87ce9aee-e5e9-4644-8b31-5ec91fc1461f)

[github-actions]

Test IDs

Added (7)

• column-change-references
• column-refs-{…}-{…}
• column-refs-{…}-{…}-after-{…}
• column-refs-{…}-{…}-definition
• column-refs-{…}-{…}-lines
• column-refs-{…}-{…}-more
• column-refs-{…}-{…}-toggle

{…} is a part filled in at run time. The full tree: docs/testing/TEST_IDS.md.