agentleFS
fourteenwmfourteenwm/ppc-ai-skills/mutation-safety/SKILL.md

mutation-safety

Skill72 starsChanged 4 months agoMITView on GitHub

MANDATORY safety system for ALL Google Ads mutations AND destructive Google Sheets writes. Auto-invoke whenever ANY agent, skill, or script attempts to modify Google Ads accounts or overwrite Google Sheets data. Enforces two-step approval to prevent accidental changes.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 9 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    ALL mutations MUST use exact matching for identifiers to prevent accidental modifications.

    line 38
  2. Distinctive:

    Default to exact match for all mutations (no flag needed)

    line 50
  3. Distinctive:

    Before executing any mutation:

    line 62
  4. Distinctive:

    If an account is not in the known accounts list, STOP and ask

    line 64
  5. Distinctive:

    When asked to modify "this account," confirm which specific account is meant

    line 65
  6. Distinctive:

    When asked to modify "all accounts," clarify the exact scope before proceeding

    line 66
  7. Distinctive:

    If your system uses approval codes or confirmation tokens:

    line 79
  8. Distinctive:

    When presenting the dry-run preview, use this structure:

    line 114
  9. Distinctive:

    If a mutation needs to be undone:

    line 157
  10. Very specific:

    Budget changes: Use op.update with the original budget value

    line 159
  11. Very specific:

    Paused entities: Use op.update to set status back to ENABLED

    line 160
  12. Very specific:

    Renamed entities: Use op.update with the original name

    line 161
  13. Very specific:

    Added negative keywords: Use op.remove with the keyword criterion resource name

    line 162
  14. Very specific:

    Conversion action changes: Use op.update with original values (name, default_value, lookback windows)

    line 163

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Mutation Safety

Purpose: Enforce two-step approval for ALL mutations to Google Ads accounts AND destructive Google Sheets writes.

Type: Safety enforcement skill (MANDATORY — auto-invokes for all mutations)


What This Skill Deliberately Does NOT Do

  • Choose the mutation. Producing skills decide WHAT to change; this protocol governs how any change gets approved and executed. Clearing the two-step gate is control, not endorsement — an approved mutation can still be a bad idea.
  • Check content or compliance. Whether copy is verified (ad-copy-verification-standard) or targeting is lawful (fair-housing-compliance) is decided upstream; this skill only guarantees a human saw the exact change first.
  • Gate reads. Queries, reports, and audits run freely — the "NOT Mutations" list below is explicit about what needs no approval.
  • Undo for you. Revert Knowledge documents HOW to reverse each mutation type, but some operations (removed conversion actions) are permanent — which is why the dry-run must flag irreversibility BEFORE execution, not after.

Core Rules

Rule 1: No Mutations Without Two-Step Approval

Any operation that writes, updates, or deletes data in Google Ads or Google Sheets MUST follow this flow:

  1. Dry Run — Show exactly what will change, with current and proposed values
  2. User Approval — Wait for explicit user confirmation before executing
  3. Execute — Only after approval, run the mutation

Never skip the dry run. Never auto-approve.

Rule 2: Exact Match Required for All Identifiers

ALL mutations MUST use exact matching for identifiers to prevent accidental modifications.

Identifiers requiring exact match:

  • Customer IDs (CIDs)
  • Conversion action names
  • Campaign names
  • Ad group names
  • Keyword text
  • Shared set names
  • Any other named entity

Implementation rules:

  • Default to exact match for all mutations (no flag needed)
  • Pattern matching requires an explicit opt-in flag
  • GAQL queries for mutations MUST use = 'value' instead of LIKE '%value%'
  • Fuzzy/partial matching is ONLY allowed for read-only queries (reports, audits)

Why this matters:

  • "Form_Submit" and "Form_Submit_BC" are different conversion actions
  • CID 1234567890 and 1234567809 are different accounts
  • One character difference can affect the wrong entity

Rule 3: Scope Verification Before Execution

Before executing any mutation:

  1. Confirm the target account is one the user actually manages
  2. If an account is not in the known accounts list, STOP and ask
  3. When asked to modify "this account," confirm which specific account is meant
  4. When asked to modify "all accounts," clarify the exact scope before proceeding

Rule 4: Show What Will Change

The dry-run preview MUST include:

  • Target: Which account(s) and entities
  • Current state: What the values are now
  • Proposed state: What they will be after the mutation
  • Count: How many entities are affected
  • Reversibility: Whether this change can be undone (and how)

Rule 5: Never Generate Approval Codes

If your system uses approval codes or confirmation tokens:

  • The USER must provide the approval code
  • Claude/AI must NEVER generate, guess, or auto-fill approval codes
  • The approval code exists so the human controls execution

What Counts as a Mutation

Google Ads API Mutations

  • Creating campaigns, ad groups, keywords, or ads
  • Updating budgets, bids, or bid strategies
  • Pausing or enabling campaigns/ad groups/keywords
  • Adding or removing negative keywords
  • Modifying conversion action settings (values, attribution, counting)
  • Updating ad copy (headlines, descriptions)
  • Any mutate call to the Google Ads API

Google Sheets Destructive Writes

  • Clearing a sheet tab
  • Overwriting existing data
  • Deleting rows or columns
  • Replacing tab contents

NOT Mutations (No Approval Needed)

  • Reading/querying Google Ads data (SELECT queries)
  • Reading Google Sheets data
  • Writing to a NEW (empty) sheet tab
  • Appending rows to the end of existing data
  • Local file operations

Dry-Run Output Format

When presenting the dry-run preview, use this structure:

MUTATION PREVIEW
================
Target: [Account Name] (CID: XXXXXXXXXX)
Operation: [What will change]
Entities affected: [Count]

Current → Proposed:
  - [Entity 1]: [current value] → [new value]
  - [Entity 2]: [current value] → [new value]

Reversibility: [Yes/No — how to undo if needed]

Type APPROVE to execute, or CANCEL to abort.

Worked Example (Synthetic Data)

A skill proposes raising Acme Plumbing's search budget. The dry-run preview it must show first:

MUTATION PREVIEW
================
Target: Acme Plumbing (CID: 1234567890)
Operation: Campaign budget increase
Entities affected: 1

Current → Proposed:
  - Budget "Search - General": $50.00/day → $54.00/day (+8%)

Reversibility: Yes — op.update back to $50.00/day restores the original state.

Type APPROVE to execute, or CANCEL to abort.

Nothing has executed yet. If the user replies APPROVE, exactly this change runs — same account, same entity, same values. If the user replies CANCEL, asks a question, or says anything else, the account stays untouched. And the approval is typed by the user — never generated on their behalf (Rule 5).


Revert Knowledge

If a mutation needs to be undone:

  • Budget changes: Use op.update with the original budget value
  • Paused entities: Use op.update to set status back to ENABLED
  • Renamed entities: Use op.update with the original name
  • Added negative keywords: Use op.remove with the keyword criterion resource name
  • Conversion action changes: Use op.update with original values (name, default_value, lookback windows)
  • Deleted/removed conversion actions: Cannot be restored via API — these are permanent

Important: Some operations are irreversible. Always flag irreversible mutations in the dry-run preview.


Common Mistakes This Prevents

  1. Wrong account scope — Script meant for one account runs across the entire MCC
  2. Partial name match — LIKE query matches 3 conversion actions instead of 1
  3. Transposed CID — Digits swapped, mutation hits wrong account entirely
  4. Auto-approve in batch — Script processes 50 accounts without stopping for review
  5. Overwrite vs. append — Sheet write replaces existing data instead of adding to it
  6. Test vs. production — Mutation runs against live account instead of test account

When to Load Other Skills

This skill is a gate, not a workflow — producing skills route INTO it whenever they reach a write step. In this repo that includes add-account-negative-keywords and sqr-pipeline (negative-keyword uploads), dgen-automation-disable (ad-automation settings), pmax-asset-automation (campaign-level automation fixes), and budget-recommendation-calculator (when an approved recommendation is written to the account).

Load alongside:

SkillWhen
ad-copy-verification-standardWhen the mutation writes ad copy — verification gates the content, this skill gates the write
fair-housing-complianceWhen the mutation touches targeting on housing accounts — compliance gates the targeting choice, this skill gates the write

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.