fix-output

name	fix-output
description	Run the adversarial output formatting loop. Output-critic checks tables/figures against AEA style, output-fixer applies formatting fixes, re-checks until APPROVED or max 5 rounds.
argument-hint	[optional: specific table/figure path]
allowed-tools	["Read","Grep","Glob","Edit","Write","Bash","Task"]

Adversarial Output Formatting Loop

Run the output-critic and output-fixer pair in an adversarial loop on tables and figures.

Input: $ARGUMENTS — optional path to a specific table or figure. If omitted, checks all output.

Steps

1. Identify target files:

   • If $ARGUMENTS specifies a file: check that file only
   • Otherwise: check all output/tables/*.tex and output/figures/*.pdf

2. Run the adversarial loop:

   ROUND = 1
   while ROUND <= 5:
   

   a. Launch the output-critic agent on target files.

   • Agent performs two passes:
     • Pass 1: Check output .tex files against AEA style guide
     • Pass 2: Check source esttab commands in .do files
   • Agent produces report at quality_reports/output_review_roundN.md
   • Verdict: APPROVED or NEEDS_REVISION

   b. Check verdict:

   • If zero Critical AND zero Major issues: APPROVED — break
   • Otherwise: NEEDS_REVISION — continue

   c. Launch the output-fixer agent with:

   • The output review report path
   • Agent applies output-level fixes (edit .tex tables)
   • Agent applies source-level fixes (edit esttab in .do files)
   • Agent logs to quality_reports/output_fixes_roundN.md

   d. Increment ROUND

3. After the loop exits, present summary:

   • Number of rounds taken
   • Output-level fixes applied
   • Source-level fixes applied
   • Scripts that need re-running (because source-level esttab commands were changed)
   • Compilation check result
   • Remaining issues (if max rounds reached)

4. If source-level fixes were applied:

   • List each .do script that was modified
   • Tell the user: "Run ./run_all.sh '[script]' to regenerate output with corrected formatting"

Important

• The output-critic does NOT edit files — it only produces a report
• The output-fixer NEVER changes coefficient values, standard errors, or any numeric content
• Output-level fixes to .tex files are temporary — they get overwritten when the pipeline re-runs
• Source-level fixes to .do esttab commands persist across re-runs
• Maximum 5 rounds
• References manuscript/aea_style_guide.md for all formatting standards