Claude Skill

bounded-growth-loop

Design a small recurring product, content, SEO, analytics, or go-to-market review loop with evidence checks, durable state, explicit stop rules, and approval-gated outputs. Use before scheduling recurring growth work or turning a repeated marketing task into an agent loop.

LLM Mart · 0 points · 3 views 0 listing impressions 0 install-command copies
Download markoblogo-abvx-agent-skills-skills_bounded-growth-loop-37a7146.zip · 2 KB

Install

skills CLI npx skills add https://github.com/markoblogo/abvx-agent-skills/tree/main/skills/bounded-growth-loop
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install markoblogo-abvx-agent-skills@llmmart
Git git clone https://github.com/markoblogo/abvx-agent-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole markoblogo/abvx-agent-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Bounded Growth Loop

Turn a proven recurring review into a safe, useful loop. The default output is a report or draft, never an unreviewed external action.

Use When

  • a content freshness, SEO, funnel, partner, activation, or market-positioning review repeats on a cadence;
  • a team wants a weekly or monthly operating rhythm without a full marketing runtime;
  • a recurring agent task needs state, deduplication, evidence checks, and a clear human gate.

Loop Contract

Specify all of these before scheduling:

  • cadence and trigger;
  • action condition, distinct from the check itself;
  • one intended outcome and bounded scope;
  • source and product context inputs;
  • self-check and evidence threshold;
  • durable state, dedupe key, cooldown, and run log;
  • proposal output location;
  • stop, error, and escalation rules;
  • approval boundary and rollback/discard path.

Workflow

  1. Read docs/product-context.md or the repo's equivalent.
  2. Select one narrow loop; begin report-only or proposal-first.
  3. Match cadence to signal speed and sample size; do not create daily noise for slow signals.
  4. Define the condition that authorizes a recommendation or draft.
  5. Retain the report, evidence, and state outside ephemeral chat.
  6. Run a verifier before proposing action; distinguish tracking defects, seasonality, and real changes.
  7. Send external copy, publishing, scheduling, spend, pricing, outreach, or product mutations through explicit approval.
  8. Review whether the loop is still useful; pause it when it creates noise or no one acts on its output.

Default Levels

  • L1 report-only: reads and summarizes with no external mutation.
  • L2 proposal-first: creates reviewable drafts or PRs behind a human gate.
  • L3 governed: only after durable state, verifier evidence, budget limits, run log, rollback, and controlled settlement exist.

Guardrails

  • Do not auto-publish, auto-send, shift ad spend, alter pricing, or contact people.
  • Do not act on sparse, stale, personally identifiable, or unverified data.
  • Do not treat a dashboard movement as causation; record alternative explanations.
  • Keep the loop product-specific and delete it when it no longer earns its operating cost.

Pair With

  • product-context for approved product language and boundaries;
  • loop-readiness-review for readiness classification;
  • bounded-evaluation for the verifier;
  • reversible-agent-task and social-publishing-gate for proposal and publication settlement.

Provenance

Adapted from coreyhaines31/marketingskills recurring-loop structure, aligned with ABVX proposal-first execution and loop-readiness gates rather than a marketing automation runtime.

Files (abvx-agent-skills)
  • agents
    • openai.yaml 234 B
      Loading…
  • SKILL.md 3 KB
    ---
    name: bounded-growth-loop
    description: Design a small recurring product, content, SEO, analytics, or go-to-market review loop with evidence checks, durable state, explicit stop rules, and approval-gated outputs. Use before scheduling recurring growth work or turning a repeated marketing task into an agent loop.
    license: MIT
    metadata:
      abvx_status: experimental
      abvx_origin: adapted
    ---
    
    # Bounded Growth Loop
    
    Turn a proven recurring review into a safe, useful loop. The default output is a report or draft, never an unreviewed external action.
    
    ## Use When
    
    - a content freshness, SEO, funnel, partner, activation, or market-positioning review repeats on a cadence;
    - a team wants a weekly or monthly operating rhythm without a full marketing runtime;
    - a recurring agent task needs state, deduplication, evidence checks, and a clear human gate.
    
    ## Loop Contract
    
    Specify all of these before scheduling:
    
    - cadence and trigger;
    - action condition, distinct from the check itself;
    - one intended outcome and bounded scope;
    - source and product context inputs;
    - self-check and evidence threshold;
    - durable state, dedupe key, cooldown, and run log;
    - proposal output location;
    - stop, error, and escalation rules;
    - approval boundary and rollback/discard path.
    
    ## Workflow
    
    1. Read `docs/product-context.md` or the repo's equivalent.
    2. Select one narrow loop; begin report-only or proposal-first.
    3. Match cadence to signal speed and sample size; do not create daily noise for slow signals.
    4. Define the condition that authorizes a recommendation or draft.
    5. Retain the report, evidence, and state outside ephemeral chat.
    6. Run a verifier before proposing action; distinguish tracking defects, seasonality, and real changes.
    7. Send external copy, publishing, scheduling, spend, pricing, outreach, or product mutations through explicit approval.
    8. Review whether the loop is still useful; pause it when it creates noise or no one acts on its output.
    
    ## Default Levels
    
    - `L1 report-only`: reads and summarizes with no external mutation.
    - `L2 proposal-first`: creates reviewable drafts or PRs behind a human gate.
    - `L3 governed`: only after durable state, verifier evidence, budget limits, run log, rollback, and controlled settlement exist.
    
    ## Guardrails
    
    - Do not auto-publish, auto-send, shift ad spend, alter pricing, or contact people.
    - Do not act on sparse, stale, personally identifiable, or unverified data.
    - Do not treat a dashboard movement as causation; record alternative explanations.
    - Keep the loop product-specific and delete it when it no longer earns its operating cost.
    
    ## Pair With
    
    - `product-context` for approved product language and boundaries;
    - `loop-readiness-review` for readiness classification;
    - `bounded-evaluation` for the verifier;
    - `reversible-agent-task` and `social-publishing-gate` for proposal and publication settlement.
    
    ## Provenance
    
    Adapted from `coreyhaines31/marketingskills` recurring-loop structure, aligned with ABVX proposal-first execution and loop-readiness gates rather than a marketing automation runtime.
    
  • SKILL_CARD.md 1.4 KB
    Loading…

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.