Output processing

Aisman can detect generated output files and help you process them in a controlled way.

Open the output list

When new output files are detected, use the New detected AI outputs panel.

• You can review each file name, session label, issue count, and title.
• You can open an output file in a read-only overlay for inspection.
• You can use selection checkboxes for batch processing.
• Files with known processing errors are highlighted so you can re-check them quickly.

Running files vs ready files

Newly detected files appear in the list before the AI agent has fully finished writing them.

• If a file is still in progress, the row shows a Running label (with an activity icon) instead of processing buttons.
• A file is considered still running until both conditions are true: step.processing.end is set and all issues[].result values are populated.
• Once the file is finalized, normal processing actions become available.
• Files are re-checked when their content changes, so the row state can update without a full page reload.

If you are using Workflow Step 4, the same detection pipeline is used there as well, so you can keep monitoring in one place without duplicate polling behavior.

When processing succeeds, Aisman uses the same archive behavior from both entry points and moves the file into outputs/processed.

Polling compares file names and finalization state, not just file count. This avoids missing updates when one file is moved and another appears between polling intervals.

Before processing

Open each file and quickly review the result text.

Look for signs that work was postponed unexpectedly, for example:

• “skipped”
• “not touched”
• “deferred”
• “not implemented in this step”

If you see these phrases without a real blocker, run a follow-up prompt and regenerate the file before processing.

Also check that the file looks like a valid Aisman output file for the current project and session.

Processing options

You can:

• Process all detected files
• Process selected files only
• Process a single file from its row
• Ignore selected files
• Do not process any files

Ignore selected is enabled only when at least one row is selected.

Use selected processing when you have a clear reason to postpone specific files.

When processing a single row out of sequence, Aisman can ask for confirmation before continuing.

For files from the same session, processing order matters:

• Earlier step files in that session must be processed first.
• Workflow Step 4 shows a guided redirect when an earlier same-session step is still pending.
• In Project output rows, out-of-sequence processing is blocked to prevent invalid imports.

Live progress during processing

When you process outputs, Aisman shows a full-page loading overlay.

• The message updates immediately with the current file name.
• For longer files, the message can include per-issue progress (issue X of Y).
• Progress updates are streamed from the server while processing runs.

When processing from Workflow Step 4, you can also follow a checklist-style progression:

1. Request sent to coding agent.
2. Matching output file detected.
3. Output file updated with issue results.
4. Agent completion confirmed.
5. File validation and process decision.

If a task reaches a state that needs your attention, Aisman can show a single consolidated prompt for one or more files and take you directly to Workflow Step 4.

Partial success behavior

If a batch contains both successful and failed files:

• Successful files are moved forward in the flow.
• Failed files stay in New detected AI outputs.
• Remaining files are not cleared automatically when unresolved errors are still present.

This lets you continue without losing context for files that still need fixes.

File-level and issue-level errors

If processing fails for specific issues in a file:

• The file row is marked as an error row.
• Opening that file shows an error panel with issue number and message details.
• Error details are stored in the output JSON (step.errors) so they remain visible after reload.

Rows with server-returned processing errors remain processable after you fix the source problem. They are not locked the same way hard client-validation failures are.

For warning-only results, you can choose to fix first or continue processing when your project settings allow it.

If all detected files are still in a running/processing state, the global unprocessed-output guard dialog is suppressed until user action is actually needed.

Processing and input file pairing

When Workflow generates output, Aisman also writes a matching input context file in .aisman/inputs/ (for example in_000_096-11.md).

• This helps resume context after agent/chat compaction.
• After output is processed or ignored, matching in_*.md files are removed automatically.

For documentation sessions, output files can also include source-file metadata under step.help:

• step.help.issuesIncluded: issue IDs covered by the documentation update.
• step.help.inputFiles: help source files used for that update (for example help_session096_260523_122540.json).

When a documentation-session output is processed successfully, Aisman moves those referenced help source files from .aisman/help/sessionX/ to the matching .aisman/help/sessionX/processed/ folder.

Documentation source generation behavior

When you generate help source content in documentation workflows:

• Aisman builds the source from completed issues that have not yet been documented.
• The generated file size is capped by MaxSourceFileSizeKb.
• The size cap uses your effective per-user setting when a per-user value exists.
• The source file is assembled incrementally; once the size cap is reached, generation stops at the current issue boundary.

If you cancel documentation generation:

• The generation progress popup closes immediately.
• The temporary generated help source file is removed.
• Documentation-status tracking for that canceled run is released so those issues can be included again later.

If help-instruction synchronization fails while preparing generation:

• Aisman shows a warning and exits the preparation flow instead of leaving the progress state hanging.
• You can correct the instruction file state and start generation again without reloading the dashboard.

Temp-number conflict handling

If processing detects temporary issue-number conflicts, Aisman can return a repair response and guide you through retry.

• Conflict checks run before import to detect duplicate negative temp numbers with mismatched issue identity.
• Repair updates can include catalog.json numbering and pending output-file temp-number rewrites.
• Retry should process the same originally requested files, not all unprocessed files.

After repair, re-run processing from the same queue context.

Domain and computer requirements

In domain-based temp-number mode, output and catalog metadata must stay aligned:

• catalog.json includes domainId and computer context.
• Output files include root domainId matching the active catalog domain.
• Missing or mismatched domain metadata can block processing until fixed.

If auto-detection cannot determine the computer name, Aisman can prompt for a selectable or new unique computer name before continuing.

Input/output issue consistency checks

Validation compares output content with its matching input file.

• Pre-existing input issues must all be present in output issue entries.
• Missing required input issue IDs are treated as hard validation errors.
• This guard prevents partial imports that silently drop scope from the original request.

Built-in Fix flow

Use the Fix action on an output row to open an assisted correction dialog.

• The dialog supports direct re-check without leaving the list.
• You can open file content from the same dialog, edit, save, and re-check.
• The generated instruction text updates when options (for example check-in message request) are toggled.
• The last interacted row stays highlighted so you can keep visual context.
• If re-check finds no remaining problems, Aisman can switch out of the error-summary state and show the normal all-fixed outcome instead of leaving an empty issue list behind.

After fixing, use re-check to refresh validation state without rebuilding your full output queue.

Auto-processing rules

If auto-processing is enabled in settings, Aisman can process ready files automatically when validation conditions are satisfied.

• Files with blocking issues stay in the queue.
• Earlier blocked files can prevent later files from auto-processing out of order.
• You can always process manually after review.

In workflow runs, successful processing can return you directly to the issue-selection step so you can verify results and continue the next pass.

When a file is processed while a related Workflow Step 4 tracker exists, Aisman keeps you in the Workflow flow so you can close that tracker and continue to the next step in sequence.

Check-in message generation

After processing, generate a checkin message and copy it to your clipboard.

In Workflow Step 4, Get checkin message copies the same generated message directly to your clipboard without opening the modal flow first.

For mixed files (existing issues plus newly created issues), generated checkin messages use resolved project issue numbers after processing, not temporary placeholders.

Successfully processed files are marked as processed to prevent accidental re-processing.

The processing flow is also what picks up files created directly by an AI coding agent, so you can return to this panel even when the session did not start from a generated prompt.

Recommended flow

1. Review output text for deferred work wording.
2. Process the file.
3. Resolve highlighted error files (if any) and reprocess only those files.
4. Generate and copy the checkin message.
5. Commit/check in.
6. Re-run only when needed.

Tips

• If no rows are selected, Process selected prompts you to choose rows first.
• Output handling can block normal navigation until you explicitly decide how to handle detected files.
• In preview mode, the output viewer includes line-oriented navigation context so you can match validation messages faster.