Debug App

Plugin check: Run node "${PLUGIN_ROOT}/scripts/check-version.js" - if it outputs a message, show it to the user before proceeding.

📋 Shared instructions: shared-instructions.md — read first.

Debug App — Monitor & Fix

Monitor the running app through the project-local .powernative/metro-logs/ files written by metro.config.js, detect runtime and bundle errors, and fix them autonomously by editing the affected files (or routing to the right skill when the fix belongs in a domain like Dataverse schema or auth registration). For silent failures, inject temporary console.log statements at data-path boundaries, read only newly appended log bytes, then clean the traces after the root cause is fixed. Modeled on the upstream app-debugger.agent.md pattern — foreground loop, bounded polling, configurable clean-check/timeout exits, and optional watch-only operation.

Dev-client limitation: the standalone dev client sends app/runtime logs, React errors, host diagnostics, and Metro bundler output through Metro. The template's metro.config.js writes Metro terminal output and HTTP bundle failures into .powernative/metro-logs/; there is no separate device log source. Host diagnostics include strings such as [AuthProvider] MSAL init failed:, [bridge] fetch THREW for, [bridge] HTTP <status> for, [addAadAppToConnectionAcl] failed HTTP <status> for connection, [useConnectionRefs] could not verify connection ACLs; treating existing connections as setup-required, and [PAHost][ErrorBoundary] Unhandled JS error:.

Subcommands (parsed from $ARGUMENTS)

Monitoring options

Options may follow the default command, a symptom, or status:

Examples:

/debug-app --port 8082 --platform ios
/debug-app "orders screen is empty" --cycles 10 --timeout 15m
/debug-app --no-fix --timeout 30m
/debug-app status --platform android
/debug-app status --working-dir ../my-mobile-app

Argument parsing:

1. Parse quoted text as one symptom value and parse recognized flags wherever they appear.
2. The first reserved token (status, stop, help, --help, -h, version, --version) selects the subcommand. stop, help, and version do not accept monitoring options. status accepts only --working-dir, --port, and --platform; reject symptoms, --cycles, --timeout, and --no-fix because it does not enter the loop.
3. After removing recognized options and their values, any remaining non-reserved text is the symptom and enables symptom mode.
4. Reject unknown flags, duplicate flags, missing values, invalid numbers, unsupported platforms, or more than one free-text symptom. Print the valid forms and exit without monitoring.
5. Normalize platform to lowercase. Convert timeout to timeoutSeconds; require 30 <= timeoutSeconds <= 3600. For monitoring and status, resolve workingDir to an absolute path from --working-dir or the current shell directory. Do not search parent directories. Require package.json and metro.config.js at that root; otherwise print the invalid path and stop before reading or writing project state. After validation, cd to workingDir once and reset it from the resulting absolute $PWD; every relative project path and command below runs from that directory.
6. Initialize:

   workingDir=<absolute app root>
   portFilter=<number|none>
   platformFilter=<ios|android|none>
   targetCleanCycles=<number, default 3>
   timeoutSeconds=<number, default 300>
   noFix=<true|false, default false>
   monitorStartedAt=<current ISO timestamp>

For help / --help / -h, print the subcommands and monitoring-options tables and exit.

Early-return subcommands:

• stop: if received while this foreground loop owns the conversation, clean up any injected traces using Phase 0.5.5 and exit at the next safe boundary. For a standalone stop invocation, report that no loop is active. Never stop Metro or delete .powernative/debug-app/ state.
• version / --version: read ${PLUGIN_ROOT}/.plugin/plugin.json, print <name> <version>, and exit without resolving a project or writing state. If the manifest is missing or malformed, report that the plugin version is unavailable and exit.
• status: validate workingDir, then execute only Phase 0.0 discovery through the status branch below. Do not run project preflight, create state files, ask for a session, initialize a baseline, or enter the monitor loop.

Tip — "play around then debug": Metro persists recent app output in .powernative/metro-logs/ even across chat/editor restarts. If something weird just happened, keep using the app normally, then run /debug-app or /debug-app "<what you saw>". The first cycle reads the latest persisted log window; subsequent cycles read only bytes appended after the saved cursor.

Core Principles

• Foreground autonomous loop — Once started, this skill owns the conversation until targetCleanCycles consecutive clean polls confirm the app is healthy, timeoutSeconds elapses, the user types stop, or the escalation rule trips. Do not run other skills concurrently — they'll queue behind the loop.
• Watch-only means no project mutation — With --no-fix, do not edit project source/configuration, inject diagnostic logs, install packages, regenerate schemas, switch accounts, or invoke a mutating skill. Continue to advance .powernative/debug-app/ cursor/audit/health state, classify errors, and provide the fix/handoff that would have been used.
• Run AFTER the app is loaded — Metro must be running through npm run dev and the simulator/device must have the app open. Phase 0 verifies that a live .powernative log exists; the skill stops cleanly if no app is detected.
• Native-only runtime target — The app must be loaded in a native dev client on a device or simulator; .powernative/metro-logs/ is the authoritative log source for that native session.
• No web or direct Metro probes — Do not use React Native Web, browser automation, curl, fetch, WebFetch, or any direct request to a Metro/localhost endpoint for runtime diagnosis. Read only the .powernative log and source files.
• No screen-by-screen verification — Do not crawl routes or validate every screen. In symptom mode, focus only on the user-reported workflow and the terminal/source evidence needed to diagnose it.
• One fix at a time — Fully resolve one issue (context → fix → type-check → reload → re-poll) before starting the next. No batching.
• Working-dir state — Debug audit state lives in .powernative/debug-app/: fixes.md, unresolved.md, injected-logs.md, health.json, and metro-cursor.json. Metro logs live in the already-ignored .powernative/metro-logs/. Both survive chat/editor restarts without binding state to a specific agent host.
• Defense-in-depth redaction — The Metro logger removes credential-like lines before writing .powernative, but /debug-app must independently minimize and sanitize every value before persisting it to .powernative/debug-app/. Never copy raw response bodies, record objects, tokens, headers, trace payloads, or absolute home-directory paths into debugger state.
• The port is the log identity — the dev-server port is the number the QR encodes, the device dials, and this skill verifies. Liveness is a socket probe (does the log's PID still hold that port?), never terminal scrollback. A port-taken status means the log belongs to a dead session and must not be diagnosed.
• Never fix history — the log is a record of the past, so an error in it is not proof of a current problem. Errors found in the baseline window must pass the Phase 0.2.1 supersession check before any code is edited. Editing working code to chase an already-resolved error is a worse outcome than reporting nothing.
• First-party native packages are an immutable boundary — a frame under node_modules/@microsoft/power-apps-native-* is evidence, not proof, of package ownership. First rule out invalid app inputs, unsupported configuration, and misuse of the package's documented API. Once the evidence confirms a defect inside an installed @microsoft/power-apps-native-* package, do not apply a customer-project workaround: do not edit node_modules/, create a patch-package patch or postinstall rewrite, vendor or fork package source, redirect the package through Metro/Babel/TypeScript aliases, or replace its dependency with a git, tarball, or local fork. Record a sanitized package-defect report and route the user to /report-issue.
• Host terminal APIs are optional only — If the current host already exposes Metro terminal output, it may be consulted as a low-latency convenience. Never ask the user for a terminal ID, never persist one, and never make a diagnosis from host output without advancing the authoritative .powernative log cursor too.
• Context-first diagnosis — Read memory-bank.md when present. Read power.config.json for environment, Dataverse, and connector context. Consult native-app-plan.md only when the failure concerns a planned screen, data model, connector, offline profile, or native capability; do not parse it for unrelated syntax/runtime errors.
• Reference resolution order — For Dataverse/Power Platform errors: read skills/add-dataverse/references/dataverse-reference.md first, inspect generated services/models and project context, then query mcp__plugin_mobile-app_microsoft-learn__microsoft_docs_search when behavior remains uncertain. For Expo/Expo Router/React Native errors: inspect installed versions and project code first, then use targeted WebFetch against official https://docs.expo.dev/ documentation. Use package documentation next and general web search only as a last resort.

Workflow — Task List First

Before entering the monitor loop, write a task list and keep it up to date:

- [ ] Discover valid `.powernative` Metro sessions and select one when needed
- [ ] Capture baseline log state from the selected session and save its byte cursor
- [ ] (Symptom mode only) Phase 0.5: parse symptom → ask user to navigate → inject console.logs → read new log bytes → walk data path → clean up logs
- [ ] Monitoring cycle 1: collect → classify → fix if needed
- [ ] Monitoring cycle 2: collect → classify → fix if needed
- [ ] Monitoring cycle 3: collect → classify → fix if needed
   (add cycles as needed; after <targetCleanCycles> consecutive clean cycles, use the resolved/flagged/pending symptom outcome)
- [ ] Fix: <error summary> → <inline edit | skill route>  (one task per error found)

Mark each cycle complete (clean OR fixed) before starting the next.

Phase 0 — Startup Check

Before entering the loop:

0.preflight Read project context

Follow the shared instructions before diagnosing logs:

Use workingDir from argument parsing as <working_dir> throughout this workflow. It is already validated as an app root; do not derive it again from a later phase or silently switch projects.

1. If <working_dir>/memory-bank.md exists, read its Project facts, Power Platform context, Data model, Connectors, Screens, Native capabilities, and Build history. Do not create a memory bank from /debug-app; absence is valid.
2. Read <working_dir>/power.config.json when present. Capture environmentId, databaseReferences, and connectionReferences. power.config.json is the preferred environment source.
3. Read <working_dir>/native-app-plan.md only when the supplied symptom or a classified error concerns a planned screen, Dataverse entity/column, connector, offline behavior, navigation contract, or native capability. Treat the plan as intended design, not proof of live schema/runtime state.
4. Inspect src/generated/services/, src/generated/models/, and src/generated/connectorSchemas.ts only when the affected data path uses them. Never edit generated files.

Append a compact context line to fixes.md after Phase 0.1 creates it:

[<HH:MM:SS>] Context — memory-bank=<present|absent> native-plan=<present|absent|not-needed> environment=<environmentId|none> dataverse-tables=<N> connectors=<N>

Do not resolve the environment or call Dataverse during ordinary bundle, React, or local JavaScript failures. When a current symptom/error is Dataverse or Power Platform related, perform the read-only Dataverse diagnostic sequence in D2 before handing off.

0.0 Discover and select the project-local Metro session

Use the validated <working_dir> from argument parsing. Enumerate all matching logs, newest first:

LOG_DIR="<working_dir>/.powernative/metro-logs"
ls -t "$LOG_DIR"/metro-*-pid-*-port-*.log 2>/dev/null

For every candidate, parse:

• pid and port from metro-<timestamp>-pid-<pid>-port-<port>.log;
• startedAt from the filename timestamp;
• project from power.config.json.appDisplayName, falling back to the basename of <working_dir>;
• platform by scanning only the latest 64 KiB for the most recent iOS ... Bundled / iOS Bundling or Android ... Bundled / Android Bundling line; use unknown when neither appears.

Check every recorded PID with a cross-platform Node probe: process.kill(pid, 0) means alive; treat EPERM as alive and other errors as gone. When port is numeric, verify ownership with lsof -nP -iTCP:<port> -sTCP:LISTEN -t on macOS/Linux when available, or netstat.exe -ano -p tcp on Git Bash/Windows. On Linux without lsof, use ss -ltnp when available. A candidate is valid/live only when its PID is alive and either owns the recorded port, has port=unknown, or no supported port-ownership probe exists. A probe that runs and reports a different PID is a contradiction, not an unavailable probe. Do not persist terminal IDs or depend on terminal output.

Selection rules:

• Apply portFilter and platformFilter to the valid/live candidates before selection. When no valid session matches, print the filters plus the available valid sessions and stop. If a platform-filtered candidate is unknown, do not guess; tell the user to use --port or omit --platform.
• One valid session — select it automatically.
• Multiple valid sessions — show one choice per session in this exact shape, then ask which to monitor:

  <project> — <platform> — port <port> — pid <pid> — started <startedAt>

  Use AskUserQuestion; identify choices internally by the complete logPath + pid + port, not by port alone. If metro-cursor.json points to one of the valid sessions, make that session the default choice but still ask. Never silently choose the newest session when more than one is live.
• No valid sessions — apply the failure branches below. Stale or contradictory logs are evidence for the report, never selectable choices.

status branch — return before selection: After filtering and liveness checks, read .powernative/debug-app/metro-cursor.json only when it is nonempty valid JSON with string logPath, numeric pid, and numeric-or-unknown port. Treat a missing, empty, malformed, or shape-invalid cursor as absent and print a warning; never overwrite it from status. Print every valid session newest-first using the normal session display shape, marking [saved] only when logPath + pid + port match the valid cursor. Then print at most the last 20 nonempty lines from fixes.md and unresolved.md, passing the displayed text through the persistence redactor first. Print stale/contradictory candidate counts without diagnosing their log content, then exit. When no valid session exists, print that result plus the applicable no-log/stale reason and exit.

For monitoring modes only, select a session using the rules above. After selection, set LOG_PATH, pid, port, project, platform, and startedAt from the chosen candidate. This selection is sticky for the current monitor run.

When no log exists, inspect metro.config.js and resolve the logging module from <working_dir> before choosing a failure branch:

• Current factory form: the config uses createPowerAppsMetroConfig from @microsoft/power-apps-native-host/config/metroConfig, and that exact subpath resolves from the project's installed dependencies. The factory installs project-local logging internally.
• Legacy direct form: the config directly imports @microsoft/power-apps-native-host/metro-logger, and that exact subpath resolves from the project's installed dependencies.
• Unavailable: neither configured subpath resolves, the referenced form is absent, or @microsoft/power-apps-native-host is missing from package.json. A dependency version string alone does not prove that the configured export exists.

Branch as follows:

The PID/port check prevents stale-log diagnosis: a log file can outlive its Metro process, so only the socket probe reveals that the log stopped belonging to the app under test. The explicit choice prevents a valid session on one port from being confused with another valid session in the same project.

If no .powernative log exists but the host exposes a live Metro terminal, that output may explain what is running, but do not ask for or store its terminal ID and do not enter the continuous monitor loop against it. Ask the user to restart with npm run dev so Metro config creates the log.

Record the stable source in fixes.md:

[<HH:MM:SS>] Log source — <project> <platform> — <logPath> (pid <pid>, port <port>, started <startedAt>)

0.1 Ensure state directory

mkdir -p .powernative/debug-app
touch .powernative/debug-app/fixes.md
touch .powernative/debug-app/unresolved.md
touch .powernative/debug-app/injected-logs.md
touch .powernative/debug-app/health.json
touch .powernative/debug-app/metro-cursor.json
rm -f .powernative/debug-app/symptom-state    # per-session — Phase 0.5 rewrites it if symptom mode is active

If fixes.md is empty, write a session header:

# Debug session — <date>

Append the effective invocation settings:

[<HH:MM:SS>] Monitor config — port=<any|port> platform=<any|ios|android> clean-cycles=<targetCleanCycles> timeout=<timeoutSeconds>s no-fix=<true|false>

0.1.1 Initialize health state

Write .powernative/debug-app/health.json without raw log content:

{
  "bundle": { "status": "unknown", "summary": "No evidence yet" },
  "runtime": { "status": "unknown", "summary": "No evidence yet" },
  "authentication": { "status": "unknown", "summary": "No evidence yet" },
  "dataverse": { "status": "unknown", "summary": "No evidence yet" },
  "connector": { "status": "unknown", "summary": "No evidence yet" },
  "offline": { "status": "unknown", "summary": "No evidence yet" },
  "navigation": { "status": "unknown", "summary": "No evidence yet" },
  "nativeCapability": { "status": "unknown", "summary": "No evidence yet" },
  "updatedAt": "<ISO timestamp>"
}

Allowed statuses are healthy, degraded, failed, unknown, and not-configured. Update only the affected domain after each classified signal:

Absence of evidence remains unknown; never convert an unobserved domain to healthy.

0.1.2 Persistence redaction gate

Before appending diagnostic text to fixes.md or unresolved.md, or storing a health summary, sanitize it:

1. Convert project paths to paths relative to <working_dir>. Replace other absolute home-directory paths with [REDACTED_PATH].
2. Replace complete authorization/cookie/secret headers (Authorization, Proxy-Authorization, Cookie, Set-Cookie, x-api-key, api-key, client-secret, and token headers) with [REDACTED_HEADER].
3. Replace Bearer values, JWT-like three-segment values, SAS/query secrets (sig, se, sp, sv, code, token, access_token, refresh_token, id_token, client_secret), passwords, and API keys with [REDACTED_SECRET].
4. Replace email addresses with [REDACTED_EMAIL] and GUIDs/record identifiers with [REDACTED_ID] unless the identifier is a non-sensitive local PID/port.
5. Never persist Dataverse/connector response bodies or arbitrary [TRACE] object values. Persist only status, source/config table or column names, counts, error codes, and a bounded error message.
6. Scan the sanitized result again. If any sensitive pattern remains, persist [REDACTION_BLOCKED: diagnostic omitted] instead.

Run the bundled verifier for every candidate diagnostic before writing:

printf '%s' "$DIAGNOSTIC_SUMMARY" | node \
   "${PLUGIN_ROOT}/scripts/redact-debug-diagnostic.js" \
  --working-dir "<working_dir>"

Persist only the verifier's stdout. Build DIAGNOSTIC_SUMMARY from minimal fields first; do not pass a full Metro window or response body and rely on redaction to make it safe.

Store logPath in metro-cursor.json relative to <working_dir> and resolve it against <working_dir> when reading. This gate is required even though .powernative logs are already sanitized.

0.2 Verify Metro bundled and the app is running

Telemetry checkpoint: validate_metro_session

Read a bounded baseline window and return a real byte cursor:

node - "$LOG_PATH" 262144 <<'NODE'
const fs = require('node:fs');
const [file, maxText] = process.argv.slice(2);
const maxBytes = Number(maxText);
const size = fs.statSync(file).size;
const start = Math.max(0, size - maxBytes);
const fd = fs.openSync(file, 'r');
const buffer = Buffer.alloc(size - start);
const bytesRead = fs.readSync(fd, buffer, 0, buffer.length, start);
fs.closeSync(fd);
process.stdout.write(JSON.stringify({
   cursor: start,
   nextCursor: start + bytesRead,
   truncated: start > 0,
   output: buffer.subarray(0, bytesRead).toString('utf8')
}, null, 2));
NODE

The logger initially writes port-unknown and may atomically rename that file to the numeric port. If the baseline read gets ENOENT, rediscover once before reporting failure. When the same filename timestamp and PID now exist with a numeric port, update LOG_PATH and port to that file and retry the bounded read once; do not process both names as separate sessions.

Parse output, cursor, nextCursor, and truncated; use the pid, port, and logPath resolved in Phase 0.0. Scan output:

• Most recent error-class line is SyntaxError, Unable to resolve module, transform failed, or error: Bundling failed → bundle is broken. Treat as a Step B "Import / Bundle" critical error and route through Step D immediately. Do NOT enter the steady-state loop until the bundle is healthy.
• Output contains Bundling complete / iOS Bundled / Android Bundled with no later error-class line → Metro is healthy. Proceed.
• Output contains a Metro banner (Metro waiting on, Logs for your project, or › Metro:) but no native Bundled / bundling lines yet → Metro is up but no native client has connected. Tell the user:

  Metro is running but no app is connected yet. Open the app on a device or simulator, then re-run /debug-app. Stop here.

• Output contains only the logger startup line while the selected PID/port session remains live → Metro has not emitted enough state yet. Tell the user to wait for the native URL, then re-run /debug-app; do not guess readiness.

If truncated: true, the baseline covers only the latest 256 KiB. Record that older history was omitted; do not claim the full session history was inspected.

Before initializing the cursor, pass all classifiable entries in this initial window through Step B, including JS runtime errors, React warnings, network/API failures, native errors, and host diagnostics. Do not advance past a prior error just because it is not a bundle error. This preserves the promise that users can "play around, then debug" after a symptom already occurred.

0.2.1 Supersession — never re-fix an error the log already shows resolved

The baseline window is history, not current state. It can contain errors that were already fixed, that the user resolved themselves, or that were transient. Fixing those is worse than doing nothing: it edits working code to chase a symptom that no longer exists, and the "fix" is unverifiable because the error cannot reproduce.

So every error found in the baseline window only (not in later polls) is treated as unconfirmed until checked for supersession. Scan forward from the error line to the end of the window:

Then:

• Superseded → do NOT fix. Record it and move on:

  [<HH:MM:SS>] Observed (already resolved, not fixed) — <error summary> — superseded by <evidence line>

• Not superseded → the error is live. Fix it normally through Step D.
• Ambiguous (no clear supersession evidence either way) → do NOT edit code yet. Ask the user whether they still see it, or wait for it to reappear in the next poll. An error that recurs after the cursor advances is confirmed live.

This applies to the baseline window only. Anything appearing in a later tail --cursor poll is by definition new output and needs no supersession check.

Symptom mode overrides this. If the user supplied a symptom, that is first-hand evidence the problem is current, so a matching baseline error is treated as live even when a later line would otherwise look like supersession.

0.3 Capture baseline

Telemetry checkpoint: capture_runtime_baseline

Use the output from Phase 0.2. Note the most recently bundled native platform (iOS / Android) and any recent runtime log lines. Append to fixes.md:

[<HH:MM:SS>] Baseline — last Metro activity: <1-line summary of most recent lines>

0.4 Initialize the durable byte cursor

Write .powernative/debug-app/metro-cursor.json using structured JSON:

{
   "logPath": ".powernative/metro-logs/metro-<timestamp>-pid-12345-port-8081.log",
   "pid": 12345,
   "port": 8081,
   "project": "My Mobile App",
   "platform": "ios",
   "startedAt": "2026-08-27T12:41:32.050Z",
   "monitorStartedAt": "<ISO timestamp>",
   "monitorConfig": {
      "port": null,
      "platform": null,
      "targetCleanCycles": 3,
      "timeoutSeconds": 300,
      "noFix": false
   },
   "cursor": 12345,
   "updatedAt": "<ISO timestamp>"
}

Set project-relative logPath, plus pid, port, project, platform, and startedAt from Phase 0.0. Set monitorStartedAt and monitorConfig from the current invocation, and set cursor to nextCursor from Phase 0.2. Resolve logPath against <working_dir> before file access. A new invocation always replaces the prior monitoring configuration and start time. On later invocations:

• Parse an existing cursor only when it is nonempty valid JSON with the expected field types. Treat a missing, empty, malformed, or shape-invalid file as no saved cursor, append a sanitized reset warning to fixes.md, and initialize from the selected session's bounded baseline. Never fail startup solely because prior debug state is partial.

• Same logPath, pid, and port, with the process still owning the port → reuse the saved cursor so old errors are not processed again.

• Multiple valid sessions → ask which session to monitor. Default to the saved session when it remains valid.

• A different session is explicitly selected, or the saved session is no longer valid → discard the old cursor and initialize from the selected session's latest log window.

• rotationLost: true from Step A → accept the returned reset cursor and record the file shrink/replacement in fixes.md.

Phase 0.5 — Symptom-driven setup (only when $ARGUMENTS is a symptom string)

Skip this entire phase if no symptom was provided. The standard log-polling loop alone is good at visible errors but blind to silent ones: an empty list because the connector wasn't added, a blank screen because useFocusEffect wasn't wired, blank rows because column names don't match the model. Phase 0.5 closes that gap.

--no-fix behavior: Parse the symptom and ask the user to reproduce it, but do not inject [INJECTED-TRACE] logs or edit any file. Read only new Metro output. If the symptom is silent and produces no classifiable output, write pending to symptom-state, append watch-only: traces not injected to unresolved.md, then enter the standard loop.

0.5.1 Parse the symptom

Extract three signals from the user's text:

Append to fixes.md:

[<HH:MM:SS>] Symptom — class=<class> screen=<path> entity=<service>

If no screen/entity match: keep screen=unknown / entity=unknown and proceed — Phase 0.5 still injects diagnostic logs and reads the terminal from whatever data path is most likely based on the symptom text.

0.5.2 Ask the user to navigate to the affected screen

The dev-player has no automation API for navigation. Ask the user:

"Please open the <screen> screen on the device/simulator, then reply ready."

Wait for the user to confirm before proceeding.

0.5.3 Inject diagnostic console.log statements and read terminal

Inject targeted console.log statements at the boundaries of the suspected data path so the Metro terminal reveals what's happening.

Injection sites — choose the minimum set that covers the symptom class:

Console.log injection pattern — all injected lines MUST use this exact format:

console.log('[TRACE <tag>]', <value>); // [INJECTED-TRACE]

• <tag> — short unique label for this site (e.g., items, render, service-response)
• // [INJECTED-TRACE] trailing comment on the SAME LINE — this is the cleanup grep key
• Log only primitives or a newly constructed allowlisted summary of counts, booleans, bounded error/status codes, and configured table/column names. JSON.stringify is allowed only for that constructed summary.
• Never log raw records, response bodies, arbitrary objects, field values, user-entered text, auth/data prop values, identifiers, URLs, headers, or credentials. The host logger removes credential-like lines, but it cannot prove arbitrary app data is free of sensitive information.
• Never inject inside src/generated/ — inject in the hook/screen that calls into it

Record every injection in .powernative/debug-app/injected-logs.md:

[<HH:MM:SS>] Injected [INJECTED-TRACE] at <file>:<line> — tag=<tag>

Then tell the user:

"I've added diagnostic console.log statements. Fast Refresh should apply them automatically. Navigate to <screen> and trigger the symptom (for example, scroll the list or tap the button). If the app does not refresh, reload it from the native dev-client menu. Reply done when finished."

Wait for the user to reply, then run the Step A tail --cursor procedure, persist nextCursor, and filter the returned output for [TRACE lines.

0.5.4 Walk the data path from terminal output

Use the [TRACE lines to walk the chain:

1. Screen TSX (app/<route>.tsx)

   • Find the useListData(...) / use*Data(...) call.
   • Check service-call options — a stray top: 0, an over-strict filter, a search: query bound to a never-cleared input, or orderBy on a missing column can each silently return zero rows.
   • Check any client-side .filter(...) after the data lands.

2. Data hook (src/hooks/useListData.ts or sibling)

   • Critical: the template hook has TWO mock-fallback paths:
     • Error path: service returns { error } → hook substitutes mock AND may call setError. Silent if the screen ignores error.
     • Empty-result path: service returns { data: [] } (no error) → hook silently substitutes mock. Always invisible without a [TRACE] log.
   • Detect: Grep for MOCK_ imports in the screen file. If present, mock data is wired in.
   • Confirm useFocusEffect is used (not useEffect) — useEffect won't re-run on back-navigate.

3. Generated service (src/generated/services/<Name>Service.ts)

   • If a TODO stub or file missing → route to /add-connector or /add-dataverse. Do NOT edit src/generated/.
   • If it exists and the [TRACE service-response] summary shows an error → use its bounded status/error code; 401/403 = auth issue; 404 = wrong resource name. Use separately emitted, already-sanitized host diagnostics for message context; do not add a raw error trace.

4. Generated model (src/generated/models/<Name>Model.ts)

   • Confirm field names match what the screen references. item.title vs cr3e9_title produces blank rows.

5. power.config.json

   • Confirm the datasources array contains the suspected entity / connector. If absent, npx power-apps add-data-source was never run for it.

6. Auth state (src/playerConfig.ts, app.config.js, auth.config.json, useAuth() hook)

   • 401 from the service wrapped as { error } — the [TRACE service-response] summary surfaces the status without persisting the error object or message.
   • OAuth deeplink handoff: verify app.config.js → expo.scheme matches src/playerConfig.ts → connectorOAuthRedirectUri, AND the same redirect URI is in auth.config.json and the Entra ID registration. If the app registration is missing, route the user to the Power Apps Wrap page via /set-app-registration-native.

Classify the [TRACE output:

Record the outcome in .powernative/debug-app/symptom-state (single line: resolved, flagged, or pending).

• Fix is clear and local → apply via Step D3 + D4 (type-check + reload + re-poll). After the fix, ask the user to interact with the screen again and read the terminal. If the [TRACE items] line shows N > 0, write resolved.
• Fix routes to another skill → tell the user, log to unresolved.md. Write flagged.
• No obvious cause → log a structured note to unresolved.md. Write pending and enter the monitor loop.

0.5.5 Clean up injected console.log statements

After the root cause is identified and a fix is applied (or Phase 0.5 concludes), remove ALL injected logs:

find app src -type f \( -name '*.js' -o -name '*.jsx' -o -name '*.ts' -o -name '*.tsx' \) \
   ! -path 'src/generated/*' -exec grep -nH 'INJECTED-TRACE' {} +

For each matching file, edit out the console.log(...); // [INJECTED-TRACE] lines. Verify with:

find app src -type f \( -name '*.js' -o -name '*.jsx' -o -name '*.ts' -o -name '*.tsx' \) \
   ! -path 'src/generated/*' -exec grep -nH 'INJECTED-TRACE' {} +  # must print zero matches

Clear the tracking file:

echo '' > .powernative/debug-app/injected-logs.md

Run npm run type-check once after cleanup.

Hard rule: Never leave [INJECTED-TRACE] lines in code. Clean up before marking the session done, even if the symptom is pending or flagged.

0.5.6 Re-enter the standard monitor loop

After Phase 0.5 completes, fall through to the monitor loop (Step A). Count clean cycles normally, but let Step C choose the exit from the single-line symptom-state: resolved may exit green, while flagged and pending use their non-green exits after targetCleanCycles. Never report a green result for a flagged or pending symptom.

Monitor Loop

Repeat until targetCleanCycles consecutive clean cycles, timeoutSeconds elapses, the user types stop, OR the escalation rule trips.

Before every collection cycle and immediately after any fix verification, compare the current time with monitorStartedAt. When elapsed time is greater than or equal to timeoutSeconds:

1. Clean up all [INJECTED-TRACE] lines before exit.
2. Append the elapsed time, cycle count, clean count, active session identity, and unresolved issue count to fixes.md.
3. Exit without stopping Metro:

   ⚠ Monitoring timeout reached after <duration>.
     Session: <project> — <platform> — port <port> — pid <pid>
     Clean checks: <clean>/<targetCleanCycles>
   Details: .powernative/debug-app/fixes.md

Step A — Collect logs

Telemetry checkpoint: collect_runtime_logs

Read .powernative/debug-app/metro-cursor.json and rediscover all live .powernative sessions using Phase 0.0's validation logic.

• If the saved logPath + pid + port is still valid, keep monitoring it even when a newer valid session has appeared. Do not interrupt the current run or jump ports.
• If the saved session is no longer valid and exactly one valid session remains, select that session and return to Phase 0.2 for a new baseline.
• If the saved session is no longer valid and multiple valid sessions remain, show the session choices and ask which one to monitor, then return to Phase 0.2.
• If no valid session remains, stop with the applicable Phase 0.0 failure instead of reading a stale log.

Otherwise collect newly appended bytes from the pinned session. When no bytes are already waiting, this foreground collector watches the log directory until that file changes or a full 5-second observation interval completes:

node - "$LOG_PATH" <saved-cursor> 262144 5000 <<'NODE'
const fs = require('node:fs');
const path = require('node:path');
const [file, cursorText, maxText, observationText] = process.argv.slice(2);
const cursor = Number(cursorText);
const maxBytes = Number(maxText);
const observationMs = Number(observationText);
const observedAt = Date.now();
let watcher;
let timer;
let finished = false;

function snapshot(observation) {
   try {
      const size = fs.statSync(file).size;
      const start = Number.isInteger(cursor) && cursor >= 0 && cursor <= size ? cursor : 0;
      const fd = fs.openSync(file, 'r');
      const buffer = Buffer.alloc(Math.min(maxBytes, Math.max(0, size - start)));
      let bytesRead;
      try {
         bytesRead = fs.readSync(fd, buffer, 0, buffer.length, start);
      } finally {
         fs.closeSync(fd);
      }
      return {
         cursor: start,
         nextCursor: start + bytesRead,
         rotationLost: start !== cursor,
         truncated: start + bytesRead < size,
         fileMissing: false,
         readError: null,
         observation,
         observedMs: Date.now() - observedAt,
         output: buffer.subarray(0, bytesRead).toString('utf8'),
      };
   } catch (error) {
      return {
         cursor,
         nextCursor: cursor,
         rotationLost: false,
         truncated: false,
         fileMissing: error && error.code === 'ENOENT',
         readError: error && typeof error.code === 'string' ? error.code : 'UNKNOWN',
         observation,
         observedMs: Date.now() - observedAt,
         output: '',
      };
   }
}

function finish(observation) {
   if (finished) return;
   finished = true;
   if (watcher) watcher.close();
   if (timer) clearTimeout(timer);
   process.stdout.write(JSON.stringify(snapshot(observation), null, 2));
}

const initial = snapshot('initial');
if (initial.readError || initial.rotationLost || initial.truncated || initial.output) {
   process.stdout.write(JSON.stringify(initial, null, 2));
} else {
   try {
      watcher = fs.watch(path.dirname(file), (_event, changedFile) => {
         if (!changedFile || String(changedFile) === path.basename(file)) finish('change');
      });
      watcher.on('error', () => finish('watch-error'));
      timer = setTimeout(() => finish('interval'), observationMs);
   } catch {
      finish('watch-error');
   }
}
NODE

Use only the returned output for this cycle. fileMissing: true means the logger may have renamed port-unknown; rediscover sessions instead of advancing the cursor. Any other non-null readError, or observation: watch-error, is not a clean cycle: report the bounded error code and stop rather than polling rapidly. Otherwise immediately persist the current logPath, pid, port, nextCursor, and a new updatedAt to metro-cursor.json, even when output is empty or contains an error; this prevents duplicate processing after interruption and handles file replacement safely.

Count a clean cycle only when readError is null, observation: interval, observedMs >= 5000, and output is empty or contains no classifiable error. An early change observation with informational output is not yet a full clean interval; process it and start the next collection cycle. If the log file changes, the PID/port check becomes contradictory, or the cursor resets because the file shrank/rotated, never count that cycle as clean.

If truncated: true, never count the result as clean. Process any nonempty output, then issue at most three additional tail calls from the returned cursor in the same cycle. If rotationLost: true, record an explicit rotation warning in fixes.md and require a fresh full observation interval before incrementing the clean counter. If data remains truncated after four chunks, record a backlog warning and continue next cycle rather than consuming unbounded context.

In the new output, surface as classifiable signal and update the matching health.json domain:

• Runtime ERROR / WARN / LOG prefixes
• Host diagnostic lines with prefixes like [PAHost], [bridge], [AuthProvider], [AuthContext], [useConnectionRefs], [useConnectionSetup], [addAadAppToConnectionAcl], [PAHost][ErrorBoundary]
• Stack frames (at <fn> (<file>:<line>:<col>))
• Bundle-class errors (Unable to resolve module, SyntaxError, transform failed) — re-classify as Step B "Import / Bundle" Critical and route through Step D
• HTTP method + status from Metro's request log (e.g., "GET /index.bundle?platform=ios&dev=true ..." 500 -) — non-200 on .bundle is a bundle/transform failure; non-2xx on connector / Dataverse hosts feeds Step B "Network / API"
• Lines containing Bundling complete / iOS Bundled / Android Bundled are informational — log to fixes.md at debug volume but do NOT classify as an issue
• [TRACE prefixed lines from injected trace statements — classify under the symptom walk (Phase 0.5.4), not as errors

Interpretation rule for host diagnostic lines:

• Treat as classifiable signal only when lines match emitted host strings such as:
  • [AuthProvider] MSAL init failed:
  • [AuthProvider] Intune enrollment failed:
  • [AuthContext] acquireTokenSilent failed for scopes:
  • [AuthProvider] Intune unenroll failed:
  • [bridge] unhandled plugin call
  • [bridge] fetch THREW for
  • [bridge] HTTP <status> for
  • [addAadAppToConnectionAcl] failed HTTP <status> for connection
  • [addAadAppToConnectionAcl] error:
  • [useConnectionRefs] could not verify connection ACLs; treating existing connections as setup-required
  • [useConnectionRefs] Failed to load connections:
  • [useConnectionSetup] could not grant connection ACL: missing Power Apps token or user OID
  • [PAHost] getConnectorToken: acquireToken threw for apiId="...":
  • [PAHost] getConnectorToken: acquireToken returned null for apiId="..."
  • [PAHost] getDataverseToken: acquireToken threw for orgUrl="...":
  • [PAHost] getDataverseToken: acquireToken returned null for orgUrl="..."
  • [PAHost][ErrorBoundary] Unhandled JS error:
  • [PAHost][ErrorBoundary] Error stack:
  • [PAHost][ErrorBoundary] Component stack:
• Treat as informational when lines are lifecycle/status-only, such as bridge setup/ready, token acquisition start/success, bridge registration, and connection-setup screen visibility.

Step B — Classify each new log entry

Telemetry checkpoint: classify_runtime_failures

Apply the 8-category table. Treat each unique stack trace / error message as one issue.

Parsing the multi-line bundle/transform block (primary mode): Metro prints these as a banner (e.g., error: Bundling failed, iOS Bundling failed) followed by an indented block. Unlike runtime stacks, the file:line is on the first non-banner line of the block, formatted as <absolute or relative path>:<line>:<col>. There are usually no at <fn> stack frames. Example to recognize:

iOS Bundling failed 412ms
SyntaxError: /Users/.../app/(tabs)/todos.tsx: Unexpected token (47:12)
  45 |   return (
  46 |     <YStack>
> 47 |       <Text>{title</Text>
     |             ^
  48 |     </YStack>
  49 |   );

Take app/(tabs)/todos.tsx:47:12 as the fix site. The recipes in D3.1 below operate on this format.

Ignore known-noisy lines:

• Require cycle: warnings from Metro
• VirtualizedList: You have a large list… without an associated crash
• Expo SDK informational banners (Starting Metro…, Connecting to…)
• React Navigation development warnings about non-serializable params (unless tied to a crash)
• USER_CANCELLED from pen input when the screen leaves state unchanged and does not show an error
• Host lifecycle/info lines with no failure indicator, for example [bridge] setupNativeHost: bridge ready, [PAHost] bridge registered, [PAHost] render: waiting for connection resolution (spinner)