Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/quitohooded/keel-skills/llms.txt

Use this file to discover all available pages before exploring further.

The model-delegation skill governs how Claude Code agents assign work to models and subagents. The principle is simple: assign the cheapest worker who can do the job well, keep the org chart shallow, and prevent anyone from quietly promoting themselves to a harder problem. This protects both cost and quality — an overpowered model on a trivial task wastes money; an underpowered model on a judgment task produces confident garbage.

​
Model tiers

Match the model tier to the kind of thinking the task needs, not to how important the task feels.
TierExampleUse for
Cheapest / fastestHaikuMechanical work with no strong judgment: inventories, searches, counts, simple extraction, direct comparisons, repetitive cleanup, format conversions
MidSonnetResearch, code exploration, reading repos, diagnosis, synthesis, judgment-bearing writing, and tasks that combine several sources
TopOpusReal planning, conflicts between sources, complex trade-offs, architecture, product decisions, and cross-cutting changes with real impact
Default downward. Reach for the top tier only when the task genuinely has the shape that tier is for. If you’re unsure whether a task needs the higher tier, it usually doesn’t. A task that mixes mechanical and judgment work should be split, not promoted wholesale.

​
Delegation limits

Five rules govern how agents may delegate to subagents:
  1. The cheapest tier never spawns its own subagents. If a mechanical task needs to delegate, the task was scoped wrong — return it to the parent to re-scope.
  2. Maximum depth: 2 levels (parent → subagent → one more). Deeper nesting loses context and accountability faster than it gains parallelism.
  3. No self-escalation. If a subagent decides it needs a smarter model, it does not upgrade itself — it returns to the parent with what it found and why a stronger model is warranted. The parent decides.
  4. A subagent cannot approve, confirm, or act in a hot zone. The green light always comes back to the human, or to the parent acting under a human’s green light. See Authorization Protocol for permission rules.
  5. Delegation does not replace reading the source. A summary from a subagent is an input, not the ground truth. Validate against the real files before acting.

​
The tool ladder

Reach for the lightest tool that can do the job well, and only climb when it genuinely cannot:
  • Simple public pages → a plain fetch.
  • Dynamic pages, pages behind a login, or pages that need interaction → a browser-driving tool.
  • PDFs → extract text first. Use a heavy visual/layout tool only when the layout itself carries meaning.
  • Local repos → prefer fast search (rg/grep), per-folder inventories, and selective reads before loading large amounts of context. Don’t pull a whole tree into context when a search answers the question.

​
Encapsulate repetition

If the same pattern shows up several times — the same multi-step lookup, the same cleanup, the same report — stop repeating it by hand. Turn it into a reusable tool, script, or documented procedure, then call that. Manual repetition is both a cost leak and an error source.

​
Project overrides

Projects may override the default tier mappings in AGENT_POLICY.md §4 (Model tier overrides). Read the policy before assigning models when one is present. For example: 
## 4. Model tier overrides

- All documentation generation tasks: use mid tier (Sonnet), not top tier.
- Dependency audits and version checks: cheapest tier (Haiku) is sufficient.
- Any change touching src/payments/: top tier (Opus) required.
Any such overrides are project-specific and authoritative — they take precedence over the defaults described above.
In one line: Cheapest model that can do it well; shallow delegation (max depth 2, no self-escalation, no subagent approvals); lightest tool first; script anything you do more than a couple of times.

Build docs developers (and LLMs) love

Get started for freeTalk to us