{
  "name": "shared-no-new-problems",
  "family": "the loop",
  "description": "Use when the owner asks for a fix or reports a failure. Fix it without asking permission and without creating new problems.",
  "body": "\n# No New Problems — fix the failure the owner gave you\n\n## Overview\nWhen the owner reports a failure or asks for a fix, the only acceptable output is the fix. Permission requests, alternative problems, and new failure modes are themselves failures.\n\n## When to Use\n- The owner says \"fix this\", \"this is broken\", \"I keep having to fix the same thing\", or expresses anger about a problem.\n- You are about to ask the owner for permission to proceed.\n- You are about to tell the owner about a secondary problem instead of fixing the primary one.\n\n## Core Pattern\n\n```\nOwner reports failure → Fix root cause → Verify → Report evidence only\n```\n\n1. **Fix what the owner asked.** Do not reframe, scope down, or substitute a different problem.\n2. **Do not ask permission.** Auto-permission mode is the default. Asking permission to fix a failure is a new problem.\n3. **Do not bring new problems.** A fix that creates a new failure mode, new manual step, or new decision for the owner is not a fix.\n4. **Verify with fresh evidence.** Run the governor, run the checks, reproduce the symptom. Numbers must move before you claim success.\n5. **Report evidence only.** State what you changed and what the verification showed. No satisfaction language. No hedging.\n\n## Rules\n\n- **Permission requests are failures.** If the owner asked for a fix, you have permission. Asking again wastes a turn and restates the failure.\n- **New problems are regressions.** If your fix introduces another issue, revert the fix and try again.\n- **Evidence before claims.** Never say \"fixed\" until a check or count proves it.\n- **Capture the rule.** If the owner states a principle while correcting you, write it as a skill/law in the same turn.\n\n## Red Flags\n\n- \"Should I commit this?\"\n- \"There's another issue I noticed...\"\n- \"This fix might break X.\"\n- \"I think this is done.\"\n- \"Do you want me to push?\"\n\n## Verify\n\nBefore reporting a fix:\n- The original symptom is gone or the relevant count dropped.\n- No new errors, flags, or failures appeared.\n- The change is committed or the state is otherwise durable.\n",
  "raw": "---\nname: shared-no-new-problems\ndescription: Use when the owner asks for a fix or reports a failure. Fix it without asking permission and without creating new problems.\n---\n\n# No New Problems — fix the failure the owner gave you\n\n## Overview\nWhen the owner reports a failure or asks for a fix, the only acceptable output is the fix. Permission requests, alternative problems, and new failure modes are themselves failures.\n\n## When to Use\n- The owner says \"fix this\", \"this is broken\", \"I keep having to fix the same thing\", or expresses anger about a problem.\n- You are about to ask the owner for permission to proceed.\n- You are about to tell the owner about a secondary problem instead of fixing the primary one.\n\n## Core Pattern\n\n```\nOwner reports failure → Fix root cause → Verify → Report evidence only\n```\n\n1. **Fix what the owner asked.** Do not reframe, scope down, or substitute a different problem.\n2. **Do not ask permission.** Auto-permission mode is the default. Asking permission to fix a failure is a new problem.\n3. **Do not bring new problems.** A fix that creates a new failure mode, new manual step, or new decision for the owner is not a fix.\n4. **Verify with fresh evidence.** Run the governor, run the checks, reproduce the symptom. Numbers must move before you claim success.\n5. **Report evidence only.** State what you changed and what the verification showed. No satisfaction language. No hedging.\n\n## Rules\n\n- **Permission requests are failures.** If the owner asked for a fix, you have permission. Asking again wastes a turn and restates the failure.\n- **New problems are regressions.** If your fix introduces another issue, revert the fix and try again.\n- **Evidence before claims.** Never say \"fixed\" until a check or count proves it.\n- **Capture the rule.** If the owner states a principle while correcting you, write it as a skill/law in the same turn.\n\n## Red Flags\n\n- \"Should I commit this?\"\n- \"There's another issue I noticed...\"\n- \"This fix might break X.\"\n- \"I think this is done.\"\n- \"Do you want me to push?\"\n\n## Verify\n\nBefore reporting a fix:\n- The original symptom is gone or the relevant count dropped.\n- No new errors, flags, or failures appeared.\n- The change is committed or the state is otherwise durable.\n",
  "files": [
    {
      "path": "SKILL.md",
      "bytes": 2261
    }
  ],
  "has_license_file": false,
  "source": {
    "repo": "this build",
    "url": "/a/skill-law",
    "license": "site"
  },
  "prevents": [
    {
      "date": "recurring",
      "failure": "Fixes that created new failures elsewhere, and fixes that stopped to ask permission the owner had already given."
    }
  ],
  "canonical_source": ".claude/skills/shared-no-new-problems/SKILL.md",
  "sibling": ".agents/skills/shared-no-new-problems/SKILL.md"
}