Overview - Ghost

GhostWorkspace

CI Pipeline Executions
- Filtered
Runs
Demo

Insights

Compare tasks
Analytics

Overview

Ghost

Loading workspace stats

Loading workspace insights...

Statistics interval

7 days30 days

Latest CI Pipeline Executions

Filter

Fuzzy

Sort by

© 2026 - Nx Cloud

Terms of Service Privacy Policy Changelog Status Docs Contact Nx Cloud Pricing Company @NxDevTools

GhostWorkspace

Agents
Replay
VCS
Self-Healing

Average execution time

9m

1981 CI pipeline executions

Computation time saved

31d 22h

Cache hit rate

88%

Protect cache integrity with sandboxing

Time to Green

52m 36s

Was 1h 3m, improved by 16%

Succeeded
31330
69fb623e Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
dfdf5a3a Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
21dd8cbb Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
ce32dd32 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
dee3af1c Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
f925de63 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
f82efa73 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
ca253386 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
53506294 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
cec60c87 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A
Succeeded
31330
02a00c54 Added a read-only guard that makes config writes throw in dev and CI no ref Deep-freezing config makes it immutable but not loud. Ghost's .js files and its CommonJS dependencies are sloppy-mode, where a write to a frozen object is dropped without an error, so a caller that mutates config fails silently and no test can see it. `node --use-strict` does not help: it makes only the entry point strict, and a `require()`d CommonJS module keeps its own strictness, with or without tsx. Enforcing the freeze that way would take 'use strict' directives across 1330 files, 6 of which have one today. A proxy trap throws whatever mode the caller is in. So `development` and anything starting with `test` - the environments this repo runs itself, the same set that validates strictly - now get a proxy whose set, delete, defineProperty and setPrototypeOf traps throw and name the key path. Production keeps the frozen object: it is the cheaper read, and the guard's value is in catching writes before they ship rather than after. GHOST_CONFIG_GUARD overrides either way. The two cannot be combined. A proxy over a deep-frozen target cannot return a wrapped child from its get trap, because the invariant for non-writable, non-configurable properties forbids handing back anything other than the target's own value, so a guarded tree is left unfrozen and relies on the traps. Children are wrapped lazily and memoised per underlying object, which keeps `===` stable between reads of the same subtree and terminates on cycles. Symbols and functions are handed back unwrapped, or iterators, spreads and array methods break. Only plain objects and arrays are wrapped. Reads cost more - 4.5ns to 21ns for a scalar, 120ns to 2.4µs for a spread of a config object - which is the other reason production keeps the freeze. It makes no measurable difference to the test suite: 8.50s against 9.04s over 9340 tests. One change outside the guard: config.set() now clones overrides with cloneDeep rather than structuredClone, because under the guard the value may be a proxy and structuredClone throws on those. The tests drive a deliberately sloppy-mode fixture, because a test file is no use here - vitest compiles those to ESM, which is always strict, so a write would throw with or without the guard. They stub GHOST_CONFIG_GUARD rather than inheriting it, so they do not depend on what a developer has exported. Verified against the unit, integration and e2e suites on both MySQL and SQLite, and a development boot. The guard found exactly one real violation while being built, fixed in the preceding commit: knex-migrator mutating the database config it was handed.
by Austin Bur...
A