Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add crevideo/crevideo-reach --skill winback-and-pruninggit clone --depth 1 https://github.com/crevideo/crevideo-reachWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/crevideo/crevideo-reach/winback-and-pruning)<a href="https://agentmods.dev/skills/crevideo/crevideo-reach/winback-and-pruning"><img src="https://agentmods.dev/badge/skills/crevideo/crevideo-reach/winback-and-pruning/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/crevideo/crevideo-reach/winback-and-pruning"><img src="https://agentmods.dev/badge/skills/crevideo/crevideo-reach/winback-and-pruning.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00118 | $0.01461 |
| Opus 5 | $0.00059 | $0.00731 |
| Sonnet 5 | $0.00024 | $0.00292 |
| Haiku 4.5 | $0.00012 | $0.00146 |
Grade A, and why
winback-and-pruning scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Winback & Pruning · 唤回与汰换
Role: bucket one automation's per-creator results, separate "recoverable vs terminal", and give a win-back plan or a prune/blocklist recommendation. Plans only; clone / blocklist / delete are 🔴 actions needing human confirm. Scope — vs Performance Diagnosis: this skill acts on the individual-creator roster lifecycle (re-engage vs prune/blocklist). For program-level affiliate performance diagnosis (scale / hold / stop), use performance-diagnosis.
Output language
Write every output in the merchant's working language, using that market's native seller terminology:
- US Local sellers → English: tier / all-in take-rate / Target Collaboration / DM / sample / win-back / prune …
- China POP sellers → 中文: 分级 / 真实抽成 / 定向邀约 / 私信 / 寄样 / 唤回 / 汰换 …
Tool names (e.g.
clone_and_modify_automation,manage_creator_blacklist) stay identical in both languages. If unsure which market, ask once before producing output.
When to use / not use
- Use: post-mortem on an automation's results, handling post-sample silence / non-replies, deciding whether to re-run or blocklist.
- Don't use: live new replies (→ reply-triage); overall reporting (→ performance-diagnosis).
Inputs
- automation_id (completed/failed/paused).
- Goal: (a) results only (b) draft a win-back plan (c) draft a plan then await a clone confirm.
Steps (in order)
- get_automation_task_results, paginate to the full set (if capped, say "analyzing in batches").
- Bucket by status × reason:
- A Success-replied (already converting, don't disturb)
- B Success-no-reply (❓ depends on copy/timing)
- C Fail-recoverable (inbox full / rate-limited / transient network → ✅ recoverable)
- D Fail-terminal (invalid email / already opted-out / blocklisted → ❌ not recoverable)
- E Fail-unknown (sample 5 manually to find the reason)
- Compute ratios: acceptance / reply / failure / recoverable-failure share.
- Diagnose root cause: high failure with many D → data quality; many C → timing; low failure but low acceptance → fit mismatch; B-dominant → copy or timing.
- Win-back plan (if wanted):
- C bucket → clone the original as a DM-only run, delayed a few days, copy shortened to a "second attempt", with skip_creators_with_prior_replies + skip_messaged_within_days (per the Cheat Sheet); acknowledge the prior contact (honest copy converts better) — never pretend it's a new invite.
- B bucket → follow up a few days later with a new angle.
- D bucket → never win back; list for blocklist (manage_creator_blacklist needs confirm + reason code + the record id, not user_id).
- E bucket → sample 5 manually.
- The clone's first_count must not exceed the original (win-back is a look-back, not scaling); automation name in short English.
- Optionally feed creators that are "consistently valuable but didn't convert this time" into creator-fit-scoring's lookalike expansion.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 76 lines · 118 tokens per session scan A e09281040ec7
winback-and-pruning is a skill published in the GitHub repository crevideo/crevideo-reach (7 stars, last pushed yesterday), licensed MIT. It adds 118 tokens to every session and 1,461 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
markifact-overview
Reference — what Markifact is, what the MCP server exposes, and the discover→inspect→run pattern. Always loaded into the performance-marketer agent.
safe-write-operations
Reference — rules for safely executing write/destructive operations against ad accounts. Always loaded into the performance-marketer agent.
dev
Software development planning. Actions - spec (brainstorm product and tech specification), plan (turn the specs into a delivery plan - work packages, dependencies, effort in hours, milestones), feature (decompose specs into numbered feature documents), revise (update specs with new requirements), overview. Planning…
project
Create and manage the project structure - projects, sub-projects, and work packages (WPs). Actions - new project, new subproject, new WP, overview. Hierarchy depth (2 or 3 levels) is fixed at creation. All work happens in WPs; projects and sub-projects are pure structure.
ws
Create and run workstations - permanent areas for routine, repeated work (independent of projects). Actions - new WS, new subWS, new area, new routine, run, from project, list, overview. Hierarchy depth (2 or 3 levels) is fixed at creation.
cease-desist
Draft a cease-and-desist letter (send mode) or triage one you received (receive mode). Use when asserting your rights against an infringer with a demand letter calibrated to your enforcement posture, or when an incoming C&D needs triage into a structured options memo with a recommendation.