How to use the Ops Capacity Planner agent skill

What the capacity-planner skill makes the agent do

Agent skills are folders of instructions (a SKILL.md file plus optional scripts and reference files) that an agent loads when a request matches the skill's description. This one is for ops teams that handle queued work such as tickets, cases and work items. When it is triggered, the agent is guided to produce three things: a capacity sizing at 70, 80 and 90 percent utilization against P50, P90 and P99 demand, with the probability of an SLA breach at each point; a utilization health check with a team verdict; and a 12-month quarterly hiring plan that accounts for ramp time, attrition, demand growth and when to add a manager.

The skill says its scripts are deterministic, use only the Python standard library, and make no LLM calls internally. The arithmetic is done by the scripts, not guessed by the model.

• Erlang-C sizing with shrinkage adjustment (capacity_modeler.py)
• Per-member utilization traffic lights and team verdict (utilization_analyzer.py)
• Quarterly hiring sequence with ramp, attrition and manager triggers (hiring_sequencer.py)

Install the capacity-planner skill

You can install the skill in one of two ways. For Claude Code, download the ZIP from the skill's page on this directory and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to share it with a repository. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

The skill folder is named capacity-planner and contains SKILL.md, the scripts/ folder, references/ and assets/. Keep that structure intact so the agent can find the scripts and reference files.

• Claude Code, personal: unzip into ~/.claude/skills/
• Claude Code, project: unzip into .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Prepare your data before you ask

The skill needs at least 90 days of historical demand data, exported from a work system such as Zendesk, Intercom, JSM, ServiceNow or Salesforce. It needs daily volume so it can compute P50, P90 and P99. If you only have averages, the skill's workflow says to pull the distribution first rather than size to a single average.

You also need Python 3 available if you want to run the scripts yourself. The scripts accept an --input path to a JSON file, --output set to markdown or json, a --sample flag with a built-in example, and --help.

• Python 3 installed (scripts use only the standard library)
• 90+ days of daily ticket or case volume
• Current FTE, average handle time, SLA target, working hours and shrinkage estimate
• A fill-out brief based on assets/capacity_brief_template.md is provided with the skill

Trigger the skill with natural requests

You do not need to name the skill. Describe the situation in plain terms that match its description: sizing ops capacity, building a headcount plan, modeling utilization risk, planning Q3 or annual support capacity, or designing CS coverage. The skill is also meant to apply when sustained team utilization is above 80 percent or when the team is growing more than 50 percent in 12 months.

Mention your function and the decision you need to make. Naming the source system and whether you have 90 days of data helps the agent choose the right path.

• "Size my support team for Q3 using the last 90 days of Zendesk volume."
• "Our utilization is around 88 percent and SLA is slipping. Is this a sizing problem or a process problem?"
• "Build a 12-month hiring plan for CX going from 15 to 30 agents, with ramp and attrition."
• "Give me the numbers to defend a headcount request to finance."

Read the outputs and act on them

Start with the 80 percent utilization row of the capacity model, which the skill treats as the sizing point. Then check the risk band (SAFE, WATCH, AT_RISK or CRITICAL) and the team verdict (HEALTHY, SQUEEZED, OVERLOADED or UNBALANCED). The skill says a spread of more than 30 percentage points across team members is UNBALANCED and should be fixed before hiring.

The skill's workflow also asks you to answer its forcing questions one at a time and write the answers down before committing the plan.

• Sizing point: the 80 percent utilization row
• Anyone sustained above 85 percent is flagged as a throughput-collapse risk
• Span of control: the source cites 7 ICs per manager as the trigger in the hiring sequence, and 10 or more as a coverage problem

Safety notes: scripts and network

The skill runs scripts, so review them before running them on sensitive data, as you would with any script. According to its metadata, it runs scripts but does not use the network. Your demand and team data stay in the files you pass to it, and the scripts are described as making no LLM calls. Check the SKILL.md and scripts yourself if your organisation has strict data rules.

• Runs scripts: yes
• Uses network: no
• Scripts are stdlib-only Python with no LLM calls inside them

Know when not to use it

The skill is for queued work. The SKILL.md says it is the wrong skill for project-style work, for engineering capacity (use vpe-advisor for DORA and cycle-time metrics), and for strategic multi-year workforce planning (use chro-advisor). If your bottleneck is not yet known, the source suggests running process-mapper first, because hiring around the wrong constraint wastes the hires.

• Engineering teams: use vpe-advisor
• Multi-year strategic workforce planning: use chro-advisor
• Work that is not queued: this model does not fit