debug-task
>-
Works with
--- name: debug-task description: >- license: MIT --- # moon task debugger A workflow-oriented diagnostic skill for troubleshooting moon tasks. This is not a reference manual — it guides you through a structured debugging flow so you can isolate the problem quickly. For conceptual background, see the [moon documentation](https://moonrepo.dev/docs). **Before you start:** Ask the user for the `<project>:<task>` target to debug. If they haven't provided a specific target, prompt them for it — the diagnostic flow requires a concrete target to inspect. --- ## Quick-start: 5-step diagnostic flow Work through these steps in order. Most issues resolve by step 3. ### Step 1: Inspect the resolved task configuration The first thing to check is whether the task is configured the way the user expects. moon merges configuration from multiple sources (global tasks, project config, inheritance), so the resolved result can surprise people. ```bash # Show the fully resolved task config (with inheritance applied) moon task <project>:<task> # Machine-readable version for programmatic inspection moon task <project>:<task> --json ``` **What to verify:** - `command` vs `script` — if the command contains pipes (`|`), redirects (`>`), chained commands (`&&`), or complex syntax, it must use `script`, not `command`. - `inputs` — are they too broad (`**/*` captures everything) or too narrow (missing source files)? Check `state.defaultInputs` (true = using default `**/*`) and `state.emptyInputs` (true = explicitly set to `[]`). Both keys are omitted from the JSON entirely when false, as is `state.setRunInCi`. - `outputs` — are they declared for build tasks? Missing outputs means the cache can never hydrate artifacts. In v2.3+, outputs also affect the **default `cacheStrategy`** of any task that depends on this one (see Step 4). - `toolchains` — is the correct toolchain(s) assigned? An incorrect toolchain means wrong tool versions. - `deps` — are task dependencies correct and complete? In v2.3+, each dep entry can carry a `cacheStrategy` (`hash` / `ignored` / `outputs`) that controls whether the dep contributes to this task's cache hash. If omitted, the default depends on whether the dep declares outputs. - `options` — check `persistent`, `runInCI`, `cache`, `affectedFiles`, `mutex`, `timeout`, `retryCount`, `allowFailure`, and `os`. - `env` — in v2.5+, environment variables can also be inherited from a **workspace-level `env`** in `.moon/tasks/**/*` (merged into the project's `env`, project wins), and the project can change the merge behavior via `workspace.mergeStrategies.env`. A variable with a surprising value may come from a layer outside the task. - `checks` <sup>v2.4+</sup> — shell scripts that run **before** the task. Their type determines the outcome: a `requirement` failing makes the task **fail**, all `condition` checks passing makes the task **skip**, and a `fingerprint` folds script output into the task hash. A surprising fail, skip, or cache invalidation often traces back to a check. - `tags` <sup>v2.3+</sup> — labels for grouping tasks. Affects targets like `:#quality` and MQL `taskTag` queries. If a task isn't matched by a `#tag` target you expected, check this list. - `type` — `build` (has outputs), `test` (default), or `run` (persistent) - `preset` — `server` or `utility` apply multiple option defaults at once. **Red flags:** - `command: 'eslint . && prettier --check .'` — shell syntax in `command` is a parse error in v2. Use `script` instead. - Empty `outputs` on a build task — cache will never restore artifacts. - `inputs: ['**/*']` — too broad, cache invalidates on every change. - A `persistent` task in a `deps` chain — moon produces a hard error at runtime. - `command: 'noop'` or `nop` / `no-op` — the task is intentionally a no-op and does nothing. moon treats these specially. - `runInCI: 'only'` — task runs in CI but NOT locally (common surprise). - `runInCI: 'skip'` — task is skipped in CI but relationships remain valid. - `os` set to a platform the user isn't on — the task is rewritten to a passing no-op at build time (`moon task --json` shows `command: noop` with cleared args/outputs). - `allowFailure: true` — the failure is still recorded and displayed, but the pipeline continues and moon exits successfully, so it's easy to miss. - A `condition` check present <sup>v2.4+</sup> — the task will **skip** whenever all conditions pass. A task that "never runs" may have a condition that always passes. - A `fingerprint` check present <sup>v2.4+</sup> — its script output is hashed, so volatile output (timestamps, versions) causes cache misses on every run. ### Step 2: Run with maximum verbosity If the config looks right, run the task with debug logging to see what moon is actually doing under the hood. ```bash # Debug-level logging with cache bypass moon run <project>:<task> --log debug --force # Deep debugging: reveal env vars and stdin passed to the process MOON_DEBUG_PROCESS_ENV=true MOON_DEBUG_PROCESS_INPUT=true moon run <project>:<task> --log trace --force ``` **What to look for in the logs:** - Toolchain resolution — is the right version of node/deno/bun/etc being used? - Hash generation — what sources are being hashed? - Affected status — is the task being skipped because it's "not affected"? - Process execution — what command is actually being spawned? **Visualize the execution graph** to spot dependency issues: ```bash moon action-graph <project>:<task> moon action-graph <project>:<task> --dot # DOT format (useful for agents) ``` > For all graph commands and output formats, see `references/environment-debug.md`. ### Step 3: Inspect cache state If the task runs but produces wrong results, or runs when it shouldn't, or doesn't run when it should, the cache is the likely culprit. ```bash # Inspect a hash manifest to see what inputs were hashed moon hash <hash> # Compare two hashes to see what changed between runs moon hash <hash1> <hash2> # Short-form hashes work too moon hash 0b55b234 2388552f ``` > For cache file locations, hash interpretation, and the `--force` vs `--cache off` comparison, see > `references/cache-issues.md`. ### Step 4: Diagnose the problem type Use this table to jump to the right reference: | Symptom | Likely cause | Quick check | Reference | | --------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------- | ------------------------------- | | Task doesn't exist | Inheritance not applied — check `inheritedBy` conditions in `.moon/tasks/**/*` against project's `toolchains`, `stack`, `layer`, `tags` via `moon project <name> --json` | `moon task <target> --json` | `references/config-mistakes.md` | | "Nothing to do" | `--affected` + no changes, `runInCI: false`, or `inheritedBy` mismatch (global task not inherited) | Check flags, `options.runInCI`, and `inheritedBy` | `references/decision-tree.md` | | `--affected` misses changed files <sup>v2.4+</sup> | Shallow git clone in CI — merge base can't be resolved, so diffs are inaccurate (moon now logs a warning) | Check clone depth; use full history or `--filter=blob:none` | `references/decision-tree.md` | | Task fails: "requirement check failed" <sup>v2.4+</sup> | A `requirement` check script exited non-zero, so the task refuses to run | `moon task <target> --json` — inspect `checks` | `references/config-mistakes.md` | | Task skipped, not affected/CI-related <sup>v2.4+</sup> | All `condition` checks passed, so the task was intentionally skipped | `moon run <target> --log debug` — look for "conditional checks have passed" | `references/config-mistakes.md` | | Task errors on execution | Wrong `command`/`script`, bad toolchain | `moon run <target> --log debug` | `references/config-mistakes.md` | | Stale cache (cached when it shouldn't be) | Inputs too narrow, missing `env` vars, or dep `cacheStrategy: 'ignored'` (the v2.3 default for output-less deps) | `moon hash <hash>` | `references/cache-issues.md` | | Cache miss (re-runs every time) | Inputs too broad, volatile outputs, or dep `cacheStrategy: 'hash'` propagating upstream churn | `moon hash <h1> <h2>` | `references/cache-issues.md` | | Cache miss from a `fingerprint` check <sup>v2.4+</sup> | A `fingerprint` check's script output is volatile (timestamps, PIDs), changing the hash every run | `moon hash <h1> <h2>` — look for the check hash | `references/cache-issues.md` | | Outputs not restored after cache hit | `outputs` misconfigured; or <sup>v2.5+</sup> a daemon-side archive/hydrate failure — swallowed by the main process, logged only by the daemon (a failed hydrate becomes a silent cache miss) | Check `.moon/cache/outputs/`; `moon daemon logs` | `references/cache-issues.md` | | Env var has unexpected value <sup>v2.5+</sup> | Workspace-level `env` in `.moon/tasks/**/*` merged in, or `workspace.mergeStrategies.env` changed the merge behavior | `moon task <target> --json` — inspect `env` | `references/config-mistakes.md` | | Cache behaves differently across git worktrees <sup>v2.5+</sup> | `cache.unstable_sharedWorktreeCache` shares blobs/manifests via the base checkout's `.moon/cache` | Check the setting and `MOON_CACHE_SHARED_WORKTREE_CACHE` | `references/cache-issues.md` | | New dependency cycle error after upgrading to v2.5 | Async graph building (now default) validates cycles strictly, per dependency-scope partition | Set `experiments.asyncGraphBuilding: false` to confirm | `references/decision-tree.md` | | Build re-runs on every upstream input change <sup>v2.3+</sup> | Dep using default `cacheStrategy: 'hash'` instead of `'outputs'` | `moon task <target> --json` — inspect dep entries | `references/cache-issues.md` | | Task not matched by `#tag` target <sup>v2.3+</sup> | Missing `tags` on the task, or `mergeTags` dropped them during inheritance | `moon task <target> --json` — check `tags` | `references/config-mistakes.md` | | Task hangs / pipeline stuck | Persistent task in `deps` chain (hard error in v2) | `moon action-graph <target>` | `references/config-mistakes.md` | | Task is slow | Dep chain bottleneck, no parallelism | `moon action-graph <target>` | `references/decision-tree.md` | | Task does nothing (no-op) | Command is `noop`/`nop`/`no-op` | `moon task <target> --json` | `references/config-mistakes.md` | | Task fails silently | `allowFailure: true` hiding errors | Check `options.allowFailure` | `references/config-mistakes.md` | | Task skipped locally | `runInCI: 'only'` set | Check `options.runInCI` | `references/config-mistakes.md` | | Task skipped in CI | `runInCI: false` or `'skip'` | Check `options.runInCI` | `references/config-mistakes.md` | | Mutex contention / deadlock | Two tasks share same `mutex` | Check `options.mutex` | `references/config-mistakes.md` | | Task times out | `timeout` option set too low | Check `options.timeout` | `references/config-mistakes.md` | ### Step 5: Validate the fix After making changes, verify the fix actually worked: ```bash # Bypass cache to force a fresh run moon run <project>:<task> --force # Disable cache entirely (no reads OR writes) moon run <project>:<task> --cache off # Verify the resolved config reflects your changes moon task <project>:<task> --json ``` **`--force` vs `--cache off`:** - `--force` ignores existing cache but **writes** new cache after execution. - `--cache off` disables caching entirely — no reads, no writes. > For all cache modes, see `references/cache-issues.md`. --- ## Common mistakes at a glance These are the issues that come up most often. For details and fixes, see `references/config-mistakes.md`. - **Shell syntax in `command`** — pipes, `&&`, redirects require `script`; v2 rejects these as parse errors. - **Missing `outputs` on build tasks** — cache can never hydrate artifacts. - **Overly broad `inputs`** — `**/*` invalidates cache on every change; be specific. - **Volatile outputs** — timestamps or absolute paths in build artifacts cause permanent cache misses. - **Persistent task in `deps`** — hard error; tasks named `dev`/`start`/`serve` auto-get `server` preset. - **`--affected` vs `--force` confusion** — `--affected` restricts; `--force` bypasses cache (they're opposites). - **`allowFailure: true` hiding errors** — the failure is still recorded and displayed, but the pipeline continues and moon exits successfully; check stderr at `.moon/cache/states/<project>/<task>/stderr.log`. - **`mutex` contention** — shared mutex serializes tasks; combined with deps can deadlock. - **`runInCI: 'only'`** — task silently skips when run locally (most surprising variant). - **Missing outputs flip dep `cacheStrategy`** <sup>v2.3+</sup> — a dep without `outputs` now defaults to `cacheStrategy: 'ignored'`. Downstream tasks stop invalidating on its changes; set `cacheStrategy: 'hash'` explicitly to restore the pre-v2.3 default. - **MQL tag fields on task queries are version-dependent** — in v2.3–v2.4, `taskTag=` and `tag=` (alias of `projectTag`) in `moon query tasks --query` silently matched _nothing_ (this also broke task tag glob targets like `:#tag-*`). Fixed in v2.5: `taskTag` matches the task's own tags, and `projectTag`/`tag` match the parent project's tags. On older versions, filter task tags with the `--tags` flag instead (`moon query tasks --tags quality`). - **A `checks` script silently changes task behavior** <sup>v2.4+</sup> — a `requirement` failing aborts the task, a passing `condition` skips it, and a `fingerprint` mixes script output into the hash. Inspect `checks` in `moon task <target> --json` when a task fails, skips, or re-runs for no obvious reason. - **Shallow git clone breaks `--affected`** <sup>v2.4+</sup> — a shallow clone (depth 1) prevents moon from resolving the merge base, so affected detection is inaccurate or empty. Use a full clone, or a blobless partial clone (`git clone --filter=blob:none`). - **Experiments are now on by default** <sup>v2.5+</sup> — `asyncGraphBuilding`, `asyncAffectedTracking`, and `nativeFileHashing` default to enabled. When bisecting graph, affected, or hashing oddities, disable the relevant experiment (config or `MOON_EXPERIMENT_*=false`) and compare — but also check the user's shell/CI for `MOON_EXPERIMENT_*` or `MOON_CACHE_*` overrides that silently change behavior. - **Daemon archiving/hydration failures are invisible in the main process** <sup>v2.5+</sup> — with the daemon enabled, task outputs are archived and hydrated in the background, and failures only appear in `moon daemon logs`. To rule the daemon out, re-run with `MOON_DAEMON=false`. - **Workspace-level `env` is a new inheritance layer** <sup>v2.5+</sup> — `.moon/tasks/**/*` files can define `env` inherited by all matching projects. Project values win on conflict, unless `workspace.mergeStrategies.env` says otherwise (`append`, `prepend`, `preserve`, `replace`). --- ## When to load references Each reference file covers a specific problem domain in depth. Load them only when the diagnostic flow points you there — don't load everything upfront. | Reference | When to load | | --------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `references/decision-tree.md` | When the symptom doesn't match the quick table above, or you need a systematic walk-through of all possibilities. | | `references/cache-issues.md` | When the problem is clearly cache-related: unexpected hits, unexpected misses, outputs not restoring. | | `references/config-mistakes.md` | When the task config is wrong: command vs script, inheritance bugs, presets, persistent tasks, affectedFiles, mutex, timeout, retries, runInCI variants, allowFailure, os. | | `references/environment-debug.md` | When you need to go deeper with env vars, log levels, trace profiles, or inspection tools. |
More Debugging skills
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
explore-code
lllllllama/rigorpilot-skills
Rigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated branch or worktree to transplant modules, adapt a backbone, add LoRA or adapter layers, replace a head, or stitch together meaningful low-risk migration ideas with rollback-aware records in `explore_outputs/`. Do not use for end-to-end exploration orchestration on top of `current_research`, trusted baseline reproduction, conservative debugging, environment setup, verified contribution claims, or default repository analysis.
safe-debug
lllllllama/rigorpilot-skills
Rigor Debug / Rigor Audit skill for deep learning research work. Use when the user pastes a traceback, terminal error, CUDA OOM, checkpoint load failure, shape mismatch, NaN loss symptom, or training failure and wants conservative diagnosis before any patching, with debug fixes clearly separated from research contributions. Do not use for broad refactoring, speculative adaptation, automatic exploratory patching, or general repository familiarization.

