Agent skill

paid-media-writes

Prepare, review, edit, approve, reject, or reconcile a paid-media change while preserving the human approval and provider verification boundary.

activeSelf-containedInstructions only358 words

Filed under ABM and paid.

From langchain-ai/paid-media-agent · 8 skill entries · 61 · pushed 2026-09-14

What it does when it runs

Prepare, review, edit, approve, reject, or reconcile a paid-media change while preserving the human approval and provider verification boundary.

Automated analysis of the skill and the 0 files bundled beside it. A skill’s own description is written to be selected by an agent, so it describes the job and not the dependencies.

Keys and connectors you must supply: None found.
Hosts it reaches: No third-party host appears in the skill or its bundled files.
Tool permissions it declares: No allowed-tools in the frontmatter. It only issues instructions, so there is nothing to bound.
Actions present in the files: None. Instructions only.

Our read

Computed from the artifact. Nothing here has been run, so none of it is a claim about how well the skill works — only about what it is and what it needs.

Maintenanceactive

Pushed within the last 30 days, with commits in the last 90.

CouplingSelf-contained

Nothing outside the agent is required.

Blast radiusInstructions only

No shell, no file writes, no network calls in the skill or its bundled files.

What it does to your systemsUnreviewed

The reading was not confident enough, or not yet confirmed, to publish.

Sends messages: No step found in the source
Writes records: Unreviewed
Spends credits: Unreviewed
Needs credentials: No key or connector found by pattern match

Read from source, not tested by running. Machine-read from SKILL.md at commit 0cc8109 with jev-1.13.0; labels that write, send or spend were also checked against the cited line by a second reviewer. How this is read

Substancebrief

Under 600 words, no bundled reference files.

LicenceAPACHE-2.0

Permits redistribution with attribution, per the repository.

Ask about paid-media-writes

Opens your assistant with this page's verified links already in the prompt.

Is this safe to install?Claude ChatGPT

Adapt it to my stackClaude ChatGPT

What else do I need for it to workClaude ChatGPT

Rather ask a human? Talk to Cheetah

Install it

View source on GitHub ↗

git clone --depth 1 --filter=blob:none --sparse https://github.com/langchain-ai/paid-media-agent.git /tmp/paid-media-agent
git -C /tmp/paid-media-agent sparse-checkout set "workspace/skills/paid-media-writes"
mkdir -p ~/.claude/skills/paid-media-writes
cp -R "/tmp/paid-media-agent/workspace/skills/paid-media-writes/." ~/.claude/skills/paid-media-writes/

Picked up without a restart. A project skill of the same name is shadowed by your personal one. For one repository only, swap ~/.claude/skills for .claude/skills. Claude Code docs ↗

The folder is the same in every client that implements the format — 46 of them — so if yours is not above, only the destination changes.

The skill

Source on GitHub ↗

Reproduced in full from langchain-ai/paid-media-agent/blob/0cc8109a1984377a573ed8d202b3b054db4d7b90/workspace/skills/paid-media-writes/SKILL.md, which is licensed APACHE-2.0 (repository). 358 words, 1 heading.

Paid-media writes

Use this skill whenever a request would change campaign, budget, bid, status, targeting, creative, conversion, audience, or another provider resource.

Read /skills/paid-media-wiki/write-safety.md.
Confirm the user asked for a change, not only an analysis.
Call discover_write_operations. It lists the admitted operations with their editable fields, units, risk, and the current execution gate. Never guess a mutation name or field.
Identify the target from an authorized read (for example the campaign id in a <platform>__get_campaign_performance or <platform>__list_campaigns result).
Call propose_change with the account alias, the admitted operation name, the target id, the changed fields, the reason, a measurement plan, and a reversal plan. The host reads the current value, derives risk flags, builds the typed proposal, and persists it. Nothing executes.
In one reply: write the proposal fields and risk flags exactly as returned as your message text, and call execute_change with the proposal id and revision in that same message. The runtime interrupts and shows a generic approve or reject card under your text, so the text is the reviewer's evidence. Do not ask for approval in prose and do not wait for a chat reply first. Only the runtime approval action authorizes execution; chat text does not.
On edit, the host creates a new revision; earlier approvals are invalid. Re-present the new revision.
After resume, report the receipt: verified, rejected, failed, or unknown, with the verified state, whether the provider acknowledged the call, and the reason. Do not claim success from anything except a verified receipt. A rejected receipt with a gate reason (writes_disabled, kill_switch, tool_not_released) means the operator has not released the live path; say so plainly.
If the receipt is unknown, explain that no blind retry is safe, check state with a read tool, and offer a new proposal if needed.

If execution returns proposal_changed, reload it with get_proposal. Do not resubmit a proposal that is executing or verifying. If its worker stopped, check the provider through read-only tools before proposing another change.

Never call a provider mutation directly, reveal raw ids or credentials, or suggest that a prompt can bypass the approval policy.

Other skills for the same job

Different authors, same problem. Matched on the words in the skill name, across every library in the catalogue except this one.

paid-media-ops by shalintripathi · 14
paid-media-audit by sidchaudhary · 1
mcp-media-inbox-triage by zapier · 339
earned-media-outreach by shawnpang · 331
attachment-media-handling by louisblythe · 159
social-media-strategy by thatrebeccarae · 147
social-media-management by manojbajaj95 · 101
paid-creative-ai by chadboyda · 77

Need help setting it up?

This page tells you what paid-media-writes does and what it needs. Cheetah builds the agent setup it runs inside: data, CRM, sequencing and the guardrails.

Book a call →

The directory stays free. There is nothing gated behind this.