@forzalabs/remora 2.2.1 → 2.3.0

package/CHANGELOG.md

@@ -6,6 +6,24 @@ The format is based on Keep a Changelog, and this project adheres to Semantic Ve
 
 ## Unreleased
 
+## V 2.3.0 - 2026-10-15
+
+**Upgrading from 2.2.** One change alters what an existing project produces: a `CSV` or `TXT` output now writes a header line, so every exported file of that kind gains a first line naming its columns. It is opt-out — `"header": false` on the output restores the old headerless shape — but it is on by default, so a destination that reads the first line as data needs that flag set before this version runs. Nothing else here touches an existing configuration: `remora report` is a new command, and the `.env` refusal only fires on an archive that should never have contained one.
+
+### Added
+
+- **`remora report [run-id]`, which packages a run into a `.zip` someone else can read.** A bug report is worth exactly what the reproduction is worth, so the bundle is designed backwards from what a reproduction needs rather than forwards from what is easy to collect: the usage row for the run (counts, per-stage timings, the error), that run's slice of the log file, the project configuration with its secrets removed, the environment it ran in, a one-page `report.md` that states the remora version, the node and platform, the consumer and its producers with the engine behind each source, the failure and its cause chain — and the commands that run the failing pipeline again on generated input. It defaults to the last recorded run and takes a run id for any other. **The bundle contains no executable**: the reproduction is a fenced block inside `report.md`, to be pasted or saved, rather than a `repro.sh` — an archive carrying a shell script is what a mail gateway or an endpoint scanner quarantines, and a bundle that never reaches the person who needs it reproduces nothing. Those commands bootstrap the two things a config carries from the machine it came from: a placeholder `.env` for every `{ENV_VAR}` the configuration names (`compile` refuses to run while a reference is unset, and placeholders are enough because `synth` contacts no source), and the local directories the sources point at (a missing destination fails the export AFTER the run, which reads as a second bug and is not one). Unpack, paste, and the reported failure happens again on synthetic rows. **It never compiles the project first**, unlike every other command: a bug report is wanted most when the configuration does NOT load, so the project is read leniently with `Environment.inspect()` and what will not load is collected as a finding rather than as a reason to refuse. For the same reason **no section can fail the command** — no run history, no log file, no `remora` directory at all — each one is recorded absent in the manifest with the reason, and the rest is still written; a reporter that dies on a broken project is useless precisely where it is needed. The log is sliced by run id AND by time window, because only the orchestrator's own lines carry `[usageId]` and a driver or parser line can be placed by its timestamp alone; a run whose process died before recording a finish leaves the window open to the end of the file, bounded by a record cap. Where `MONGO_URI` is set there is no local history to read at all (`UsageManager.recordsLocally()` is false), so the command degrades to the configuration, the environment and the tail of the log, and says so in the bundle
+
+- **The configuration travels as authored; the log is scrubbed of credential shapes.** The asymmetry is deliberate. A remora config is not supposed to HOLD a credential — it names one (`{PG_PASSWORD}`, resolved by `SecretManager` at the point of use) and every authentication field already accepts a reference — so `config/` is shipped byte for byte: what a dev reads there is what runs, which is the whole reason it is in the bundle, and the reproduction copies it straight into a project directory. A literal password in a source file is an authoring mistake, and rewriting it on the way out would only hide it from the one person positioned to fix it. A log is the opposite case and gets no such assumption: by the time a value reaches a log line it has been resolved, so `SecretRedactor` removes the shapes a credential has in free text — a JWT, an AWS access key id, the query string of any URL (a presigned URL carries its signature there), and `password=`/`"apiKey":`/`Authorization: Bearer` pairs — while keeping `{ENV_VAR}` placeholders verbatim, since which variable a source reads is part of the question being asked. Nothing else travels either way: only `.json` configuration files are carried, and no row of customer data is collected at any level this version can produce
+
+### Changed
+
+- **A `CSV` or `TXT` export now writes a header line.** A positional file carries values and nothing that names them, so every exported `CSV`/`TXT` file now opens with its field names — the exported ones, a field's `alias` where it has one and its `key` otherwise, in column order, quoted and delimited exactly as the rows under them so the header parses with the same reader the data does. A `hidden` field is not written and so is not named, and with a `split` **every** file of the output carries the header rather than only the first: a file that cannot be read on its own is not a file anyone can be handed. **This changes the bytes every existing CSV/TXT consumer produces**, and a destination that reads the first line as data will read the header as a row — set `"header": false` on that output to keep the old shape. The flag is `CSV`/`TXT` only: on a `JSON` or `API` output, where each record already carries its own field names, it is rejected at compile rather than ignored, so a config that asks for a header always gets one. Nothing internal changed with it — the header is composed and written at export time, not held in the dataset, so the split pass, the consumer-as-input capture, lookups and `sample` all still read a dataset of records and only records. It costs no pass over the data either: the destination drivers write the line ahead of the file they were already copying or streaming
+
+- **`remora package` and `remora deploy` now refuse an archive containing a `.env` file.** A project keeps its `.env` above `./remora` and so out of reach of the packager's walk — but that is a property of where someone put a file, and this archive is uploaded to a worker and is now also packaged into bug reports. The check makes it a property of the packager instead, with a message saying to move the file. `ConfigPackager` moved from the CLI into `packages/common` on the way, so the deployment archive and the bug report share one set of exclusions rather than two that can drift
+
+- `EnvironmentVariables` is now derived from a runtime `ENVIRONMENT_VARIABLES` list rather than written beside one, so a reader that has to enumerate the variables — the bug report's presence map, which records which are SET and never what they hold — gets exactly the names the type allows. `REMORA_MASK_IN_DEV` joins the catalogue, which the docs already named and the type did not
+
 ## V 2.2.1 - 2026-10-15
 
 ### Fixed