@stacknext/cli 0.6.0 → 0.7.0

package/skills/stacknext-debugging/references/tools.md

@@ -0,0 +1,369 @@
+# StackNext MCP tool reference
+
+All twenty tools are read-only, idempotent and restricted to the local
+collector. Tool text content is pretty-printed JSON; clients that support
+structured content also receive the same object in `structuredContent`. The
+object always contains the tool payload plus a top-level `meta` object.
+
+## Common arguments
+
+| Argument | Type | Notes |
+| --- | --- | --- |
+| `limit` | integer 1-200 | Default 50. The collector caps it at 200. |
+| `offset` | integer >= 0 | Use with `meta.result.nextOffset` when `hasMore` is true. |
+| `minDurationMs` | number 0-600000 | Inclusive threshold. |
+| `sort` | `recent`, `oldest`, `slowest` | List tools that support sorting. |
+| `q` | string, max 200 | Case-insensitive text search over the fields the tool exposes. |
+
+## Regression ids and statuses
+
+Regression ids are strict: `run-<16 hex>`, `base-<16 hex>` and
+`cmp-<16 hex>`. An id that does not match is rejected before storage is read.
+The regression tools read only the collector's configured project directory;
+they cannot read arbitrary files.
+
+- Run status: `running`, `completed`, `failed`, `interrupted` or `incomplete`.
+- Comparison status: `improved`, `regressed`, `unchanged` or `inconclusive`.
+- Severity: `high`, `medium`, `low` or `none`, separate from status.
+
+A scenario sample is `verified` only when the collector found the injected trace
+and a server span parented to the injected span id. `incomplete` means the
+telemetry could not be proven, never that a time window matched. Baselines are
+immutable snapshots and are never deleted by these tools.
+
+## Common response metadata
+
+```json
+{
+  "meta": {
+    "tool": "list_requests",
+    "generatedAt": "2026-11-02T12:00:00.000Z",
+    "project": "tasklight",
+    "timeRange": {
+      "firstEventTime": "2026-11-02T11:55:00.000Z",
+      "lastEventTime": "2026-11-02T11:59:59.000Z"
+    },
+    "result": {
+      "found": 120,
+      "returned": 50,
+      "limit": 50,
+      "offset": 0,
+      "hasMore": true,
+      "nextOffset": 50
+    },
+    "connection": {
+      "instrumentation": "connected",
+      "instances": 1,
+      "lastSeen": "2026-11-02T11:59:59.000Z",
+      "sdkVersion": "0.6.1",
+      "collectorVersion": "0.6.1"
+    },
+    "completeness": { "complete": true, "reasons": [] },
+    "notes": []
+  }
+}
+```
+
+- `result.found` is the total number of matching records in the retention
+  window, not the number returned in this page.
+- `result.hasMore` and `nextOffset` describe pagination. `nextOffset` is `null`
+  on the last page.
+- `connection.instrumentation` is `connected`, `waiting` (collector is running
+  but no app has connected) or `unknown` (health lookup failed).
+- `completeness.complete` is false when telemetry is dropped, a result is
+  paginated, no app is connected, or a trace has missing parents. Read
+  `completeness.reasons`.
+- `notes` carries short caveats, such as omitted stacks, unmeasured query
+  durations or incomplete regression correlation.
+- `get_project_summary` adds `meta.coverage` with counts (`observed`, `active`,
+  `configured`, `notObserved`, `unsupported`, `unknown`,
+  `requestsInSample`). It also returns the full `coverage` report.
+- Detail responses also include `meta.identifiers`, for example
+  `{ "traceId": "..." }`, `{ "runId": "run-..." }`,
+  `{ "comparisonId": "cmp-..." }` or `{ "snapshotId": "snap-..." }`.
+- Times in payloads are Unix epoch milliseconds. Metadata times are ISO 8601
+  strings. Regression records use their own `createdAt` timestamps and are not
+  limited by the telemetry retention window. Snapshot records use their save
+  time and sample window.
+
+## get_project_summary
+
+Arguments: none.
+
+Returns `{ health, summary, coverage, meta }`. `summary` contains total/failed/slow
+request counts, average and p95 duration, span/error/query/fetch counts, the five
+slowest requests, five recent errors (without stacks), and the first/last event
+time. `coverage` contains the sample window, status counts and one item per
+framework/adapter/custom feature. Start here.
+
+Coverage statuses are `observed`, `active`, `configured`, `not-observed`,
+`unknown` and `unsupported`. There is no coverage percentage. A `not-observed`
+or `unknown` item means evidence is missing from the current sample, not that
+the app is broken. Read `coverage.items[].limitations` before recommending
+instrumentation changes.
+
+## list_requests
+
+Arguments: `limit`, `offset`, `q`, `method`, `status`, `slow`, `minDurationMs`,
+`sort`.
+
+Returns `{ items, total, limit, offset, meta }`. Each item includes `traceId`,
+route, method, status code/class, duration, timestamp, span/error/query/fetch
+counts, `failed`, `slow` and service name.
+
+- `status` is one of `2xx`, `3xx`, `4xx`, `5xx`, `none`.
+- `method` is matched case-insensitively by the collector.
+- `slow` uses the project's `slowRequestMs` threshold.
+- `sort` defaults to `recent`.
+
+## get_request
+
+Arguments: `traceId` (32 hexadecimal characters).
+
+Compatibility alias for `get_trace`, kept for clients that use the request
+vocabulary. Returns the same detail: request summary (or `null` when no request
+root was derived), every captured span, correlated errors and database queries,
+start/end/duration and `missingParentSpanIds`.
+
+Use the span tree to find the longest span, the failing span, or the gap between
+request start and first child.
+
+## list_traces
+
+Arguments: `limit`, `offset`, `q`, `sort`.
+
+Returns `{ items, total, limit, offset, meta }`. Each trace summary includes the
+root span, service, start/end, duration, span/error/query counts, `failed` and an
+optional request summary.
+
+## get_trace
+
+Arguments: `traceId` (32 hexadecimal characters).
+
+Returns the same detail shape as `get_request`, plus an additive `diagnostics`
+array of finding references that mention this trace. This is the canonical
+detail tool; `get_request` is a compatibility alias for the same data.
+
+Each reference has `id`, `ruleId`, `category`, `severity`, `status` and `title`.
+The array is a pointer, not a second full trace: call
+`stacknext_get_diagnostic` for the finding and its bounded evidence.
+
+## list_errors
+
+Arguments: `limit`, `offset`, `q`, `route`, `name`, `includeStack`.
+
+Returns `{ items, total, limit, offset, groups, meta }`. Items include the error
+id, fingerprint, name, message, route, timestamps, trace correlation and
+`resolvedTraceId`. `groups` aggregates by fingerprint with count and first/last
+seen.
+
+Stacks are omitted by default for context efficiency. Set `includeStack: true`
+only when the list itself needs stack evidence; otherwise call `get_error` for
+one error. When stacks are omitted, `meta.notes` says so.
+
+## get_error
+
+Arguments: `errorId`.
+
+Returns `{ error, trace, meta }`. `error` includes the redacted stack and
+`resolvedTraceId`. `trace` is the full correlated trace detail or `null` when no
+trace could be resolved. Treat the correlation as exact only when the error has
+a captured span context; route/time fallback is weaker.
+
+## list_database_queries
+
+Arguments: `limit`, `offset`, `q`, `minDurationMs`, `traceId`,
+`databaseSystem`.
+
+Returns `{ items, total, limit, offset, meta }`. Each item includes redacted
+`sql`, `databaseSystem`, `operation`, `success`, `errorMessage`, `source`,
+`durationMs`, `resolvedTraceId` and `route`.
+
+`durationMs` is `null` for Drizzle queries because the Drizzle logger does not
+expose duration. `null` means unmeasured, never zero. `minDurationMs` excludes
+null-duration queries because they cannot be compared to a threshold.
+
+## find_slow_requests
+
+Arguments: `minDurationMs` (default 500), `limit`.
+
+Returns `{ items, total, limit, offset, meta }`, sorted slowest first. This is a
+convenience wrapper over `list_requests` with `sort: "slowest"` and the duration
+filter applied. Use `list_requests` when you also need method/status filters.
+
+## stacknext_list_diagnostics
+
+Arguments: `limit`, `offset`, `category`, `severity`, `status`, `ruleId`,
+`route`, `traceId`, `since`, `until`, `q`.
+
+Returns `{ items, total, limit, offset, generatedAt, engineVersion,
+ruleSetVersion, sample, counts, notes, limitations, meta }`.
+
+- `category`: `database`, `http`, `performance`, `errors` or `runtime`.
+- `severity`: `info`, `warning` or `error`.
+- `status`: `observed`, `possible` or `insufficient-evidence`.
+- `ruleId`: `n-plus-one-queries`, `slow-request`, `slow-span`,
+  `repeated-external-http`, `error-fingerprint` or `server-error-response`.
+- `since`/`until` are epoch milliseconds and filter the finding window.
+
+Findings are deterministic heuristics over the current retention window. Read
+`sample`, `notes` and `limitations` before drawing conclusions. `sample`
+includes request-trace and attribution integrity counts, so detached traces and
+ignored control-flow errors stay visible. `total` is the number of matching
+findings; `items` is the current page. Findings are already grouped by rule and
+canonical route; use `affectedTraceCount` and `evidence.traces` for occurrence
+evidence. Use `stacknext_get_diagnostic` to inspect one.
+
+## stacknext_get_diagnostic
+
+Arguments: `diagnosticId` in the form `diag-xxxxxxxx-xxxxxxxx`.
+
+Returns `{ finding, trace, meta }`. `finding` contains the rule, category,
+severity, status, route, time window, affected trace count, fingerprint, bounded
+evidence, `verification` steps and `limitations`. Evidence may include self-time
+`contributors`, per-trace `traces` occurrences, span self time and Next.js
+`nextPhase` values. `trace` is a related trace context
+bounded by the server, or `null` when the evidence aged out. Use `get_trace` for
+the complete span tree when the detail response is truncated.
+
+A diagnostic id is not a credential. It identifies a deterministic finding in
+the current sample; it can disappear when the underlying telemetry ages out.
+
+## stacknext_list_test_runs
+
+Arguments: `limit`, `offset`.
+
+Returns `{ items, total, limit, offset, meta }`. Items combine live collector
+runs with saved runs from `.stacknext/runs`. Each summary includes `id`, `name`,
+`projectName`, `status`, `createdAt`, `completedAt`, `durationMs`, scenario and
+sample counts (`samples`, `verifiedSamples`, `incompleteSamples`) and the
+captured `environment`. Live runs can be `running`; a non-completed run has
+incomplete telemetry and supports no regression verdict.
+
+## stacknext_get_test_run
+
+Arguments: `runId` (`run-<16 hex>`).
+
+Returns the full run plus `meta`: `format`, `version`, `id`, `name`,
+`projectName`, `status`, `createdAt`, `completedAt`, `command`, `environment`,
+`collector`, `scenarios`, `warnings`, `notes`, `limitations` and `stats`. Each
+scenario contains its stable `identity` (including `scenarioId`, title, file and
+project), `samples`, aggregated `findings`, notes and limitations.
+
+Each sample has its Playwright `status`, expected status, retry/repeat index,
+duration, `capturedAt`, findings and `correlation`. Correlation is `verified` or
+`incomplete` with bounded `reasons` and captured traces. Never turn an
+incomplete sample into evidence by matching timestamps or nearby requests.
+
+## stacknext_list_baselines
+
+Arguments: `limit`, `offset`.
+
+Returns `{ items, total, limit, offset, meta }`. Each baseline summary includes
+`id`, `name`, `projectName`, `createdAt`, `sourceRunId`, scenario/sample counts
+and the recorded `environment`. Baselines are immutable and are never removed
+automatically.
+
+## stacknext_get_baseline
+
+Arguments: `baselineId` (`base-<16 hex>`).
+
+Returns the baseline plus `meta`: `format`, `version`, `id`, `name`,
+`projectName`, `createdAt`, `sourceRunId`, `environment`, `scenarios`,
+`warnings`, `notes`, `limitations` and `stats`. `meta.identifiers` also carries
+`sourceRunId`. The baseline does not follow later telemetry or app changes.
+
+## stacknext_list_comparisons
+
+Arguments: `limit`, `offset`.
+
+Returns `{ items, total, limit, offset, meta }`. Each item includes `id`,
+`projectName`, `baselineId`, `candidateId`, `createdAt`, `status`, `severity`,
+`summary` counts and `environmentCompatible`. Status and severity are separate;
+`inconclusive` is not a pass.
+
+## stacknext_get_comparison
+
+Arguments: `comparisonId` (`cmp-<16 hex>`).
+
+Returns the comparison plus `meta`: `format`, `version`, `id`, `projectName`,
+`createdAt`, `baselineId`, `candidateId`, `status`, `severity`, `environment`,
+`options`, `summary`, `scenarios`, `warnings` and `limitations`.
+
+- `environment.compatible`, `environment.ignored` and `environment.differences`
+  explain the environment gate. `--ignore-environment` is a CLI option; this
+  tool only reads the saved record.
+- Each scenario includes `scenarioId`, title, file, project, `status`,
+  `severity`, `reasons`, `metrics`, and baseline/candidate sample counts.
+- Metrics include `kind`, `status`, `severity`, `baselineValue`,
+  `candidateValue`, `delta`, `percent`, `sampleCount`, optional `note` and
+  `unmeasured`.
+
+Comparisons use medians and explicit thresholds. They are not significance
+tests and do not prove causality. Read the reasons and metrics before explaining
+a result.
+
+## stacknext_list_snapshots
+
+Arguments: `limit`, `offset`.
+
+Returns `{ items, total, limit, offset, invalid, meta }`. Each item includes
+`id` (`snap-<16 hex>`), `createdAt`, `source` (`manual`, `exit`, `error` or
+`import`), `projectName`, `collectorVersion`, optional `label`, `sample`
+(`firstEventTime`, `lastEventTime`, `retentionMs`), `stats` and `bytes`.
+`invalid` counts unreadable summary files that are left on disk rather than
+repaired or deleted.
+
+## stacknext_get_snapshot
+
+Arguments: `snapshotId` (`snap-<16 hex>`).
+
+Returns `{ snapshot, config, limitations, traces, errors, databaseQueries,
+meta }`. The first 20 traces, errors and queries are included. `config` is the
+captured collector config, not current live config. `meta.identifiers` carries
+the `snapshotId`.
+
+A snapshot has no diagnostics or coverage and never updates. Use
+`stacknext_get_snapshot_trace` for one full scoped trace.
+
+## stacknext_get_snapshot_trace
+
+Arguments: `snapshotId` (`snap-<16 hex>`), `traceId` (32 hexadecimal
+characters).
+
+Returns the same trace detail shape as `get_trace`, scoped to the frozen
+snapshot, plus `meta`. It intentionally has no `diagnostics` array. A missing
+trace or snapshot returns `isError: true`; use the list/detail tools to find
+current ids.
+
+## Error responses
+
+Tool failures return `isError: true` with a short text message and no structured
+content. Typical messages:
+
+- `Cannot reach the StackNext collector at ...` — the collector is stopped or
+  the runtime file is stale. Ask the user to run `stacknext dev`.
+- `StackNext collector rejected the token ...` — the token changed after a
+  restart. Restart the MCP client or pass the current token.
+- `No request with trace id ... was found ...` — the id is wrong or the trace
+  aged out of the retention window. Use `list_requests` to find a current id.
+- `No error with id ... was found ...` — the error aged out or the id is wrong.
+  Use `list_errors`.
+- `No diagnostic with id ... was found ...` — the finding aged out or the id is
+  wrong. Use `stacknext_list_diagnostics` to find a current id.
+- `No saved test run with id ... was found ...` — the run id is wrong or the
+  record is outside the collector's project directory. Use
+  `stacknext_list_test_runs`.
+- `No baseline with id ... was found ...` — use `stacknext_list_baselines`.
+- `No comparison with id ... was found ...` — use
+  `stacknext_list_comparisons`.
+- `No snapshot with id ... was found ...` — the snapshot id is wrong or the
+  record is outside the collector's project directory. Use
+  `stacknext_list_snapshots`.
+- `No trace ... was found in snapshot ...` — use `stacknext_get_snapshot` to
+  see the traces stored in that snapshot.
+
+Do not treat an empty result as a tool failure. Empty results are successful
+responses with `total: 0` and a completeness reason. Missing regression or
+snapshot records mean the MCP server has nothing saved to read; they do not
+justify creating a record from telemetry.