AI Agent Integration

FitOps is built first as a tool you can use directly — run a command, read the output, make decisions. The agent-integration capability is an additional layer, not the primary design goal.

The Default: Coach-Style Output

When you run a FitOps command, you get readable output by default:

$ fitops analytics training-load --today

  CTL (Fitness)    48.3
  ATL (Fatigue)    61.2
  TSB (Form)      -12.9
  Form label       Overreaching — back off or rest

No flags required. No parsing. Just information you can act on.

Adding --json for Agents and Scripts

Every command accepts a --json flag that switches to structured JSON output. Use this when you want to pipe data into an AI agent, a script, or the local dashboard:

fitops analytics training-load --today --json
fitops activities list --sport Run --limit 10 --json
fitops analytics vo2max --json

When --json is used, every response includes a _meta block so the agent knows when the data was generated and what filters were applied:

{
  "_meta": {
    "generated_at": "2026-03-11T09:15:00+00:00",
    "total_count": 10,
    "filters_applied": { "sport_type": "Run", "limit": 10 }
  },
  "activities": [ ... ]
}

Values carry explicit units so an agent can reason about them without guessing:

"distance": {
  "km": 10.23
}

Token Efficiency

FitOps outputs are optimized for LLM token budgets. Compared to v0.1.0:

• 21.9% fewer tokens per activity on average
• Saves ~2,800 tokens over a 50-activity query (~56 tokens/activity)
• Old: ~1,023 chars/activity → New: ~799 chars/activity

Key optimizations:

• flags, social, and data_availability blocks are omitted entirely when empty (sparse True-only maps instead of full True/False maps)
• distance emits only km (removed redundant meters and miles)
• speed emits only km/h (removed raw m/s values)
• elapsed_time_seconds removed (use moving_time_seconds instead)

Measured across 50 real Strava activities (mix of runs and rides).

Feeding Data to an AI

Combine multiple commands and pipe the output into your AI assistant:

echo "=== Training Load ===" && fitops analytics training-load --today --json
echo "=== VO2max ===" && fitops analytics vo2max --json
echo "=== Recent runs ===" && fitops activities list --sport Run --limit 10 --json

Paste the combined output with a prompt like:

"Based on this training data, am I ready to race this weekend?"

Or use the fitops notes command to build a persistent memory layer for your agent — notes survive across sessions and can be tagged, linked to activities, and recalled by tag:

fitops notes list --tag pattern        # recall flagged patterns
fitops notes get hr-drift-march-2026   # read a specific observation

Form and Zone Labels

JSON output includes text labels alongside numbers, so an agent gets human-readable context alongside the raw values:

"form_label": "Overreaching — back off or rest"
"ramp_label": "Moderate build"

When to Use Which Mode

Situation	Use
Checking your training at the terminal	Default (no flag)
Asking an AI "how am I doing?"	--json + paste output
Scripting a daily snapshot	--json + pipe to file or agent
Browsing visually	fitops dashboard serve
Persisting coaching observations	fitops notes create

---

Setting Up FitOps as a Skill in Your AI Agent

FitOps ships a ready-to-use skill file that gives any AI assistant full context on the CLI commands, output format, and coaching patterns — without you having to re-explain it each session.

The skill file is at docs/official/fitops-skill.md in the repo, or you can grab it from the GitHub releases page.

Claude Code

Claude Code loads project-level slash commands from .claude/commands/. Drop the skill file there:

mkdir -p .claude/commands
cp fitops-skill.md .claude/commands/fitops.md

Then in any Claude Code session in that directory:

/fitops What's my current fitness level?

You can also add it as global context in ~/.claude/CLAUDE.md so it's available everywhere — useful if you use FitOps across multiple projects.

Codex (OpenAI CLI)

Codex reads AGENTS.md from the current directory (and parent directories up to repo root). Put the skill content in an AGENTS.md:

cp fitops-skill.md AGENTS.md

Or append it if you already have an AGENTS.md:

cat fitops-skill.md >> AGENTS.md

Codex will load it automatically on the next session in that directory.

OpenCode (sst.dev)

OpenCode reads instructions from AGENTS.md and rules from .rules files in the project root:

cp fitops-skill.md AGENTS.md
# or add to an existing AGENTS.md

Alternatively, create a .rules file for always-on context:

cp fitops-skill.md .rules

Cursor

Add a Cursor rule to apply FitOps context to any session:

mkdir -p .cursor/rules
cp fitops-skill.md .cursor/rules/fitops.mdc

The .mdc format is standard Markdown — no changes needed to the file. Cursor picks up all .mdc files in .cursor/rules/ automatically.

For a project where FitOps is a core tool, you can also paste the skill content into Cursor's global rules at Settings → General → Rules for AI.

Aider

Add the skill file as a read-only context file via .aider.conf.yml:

# .aider.conf.yml
read:
  - fitops-skill.md

Or pass it directly on the command line:

aider --read fitops-skill.md

Any Other Agent

The skill file is plain Markdown. You can paste it into any system prompt, context window, or instructions field:

• ChatGPT / Claude.ai: paste as a system message or first user message
• Cursor AI chat: paste into the context before asking
• Zed AI: add to your assistant instructions
• Any agent that accepts a system prompt or context file

---

The FitOps Skill File

The skill covers:

• Every command group (auth, sync, activities, athlete, analytics, workouts, race, weather, notes, dashboard)
• Full flags and output fields for each command
• How to chain commands for common workflows (fitness check, zone setup, VO2max, overtraining assessment)
• Error recovery patterns (no activities → sync, no zones → infer, etc.)
• How to interpret key metrics (TSB form zones, ACWR bands, HR drift %, WAP factors, WBGT heat flags)
• The agent memory pattern using fitops notes

← Concepts