Skip to content

Commit 88c2778

Browse files
docs(agents): clarify design patterns between context-manager and specialty replicas; add design checklist and usage tips
Why: align agent docs with clarified architecture patterns and provide a practical checklist to guide task breakdown and orchestration. 🤖 Generated with Codebuff Co-Authored-By: Codebuff <noreply@codebuff.com>
1 parent 26140c8 commit 88c2778

File tree

1 file changed

+21
-4
lines changed

1 file changed

+21
-4
lines changed

web/src/content/agents/customizing-agents.mdx

Lines changed: 21 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -21,14 +21,29 @@ Create specialized agents from scratch using TypeScript files in the `.agents/`
2121

2222
Agents adapt to your specific workflow and project needs.
2323

24-
Keep in mind you'll get the most value from agents if you see them as a means of managing your context window. Think about how you want to break down tasks and build your agents around that, as opposed to replicating engineering specialties.
25-
26-
Comparison: Context managers vs. specialty replicas
24+
Keep in mind you'll get the most value from agents if you treat them as context window managers. Design them to orchestrate and sequence work so the right files, facts, and decisions are in scope at each step. Break tasks into steps and build agents around controlling that flow — instead of trying to replicate human specialties.
2725

2826
<MarkdownTable>
29-
27+
| Aspect | Context manager agents | Specialty reviewer agents |
28+
| --- | --- | --- |
29+
| Primary job | Orchestrate multi-step workflows and manage context | Perform deep, targeted reviews or focused changes |
30+
| Best for | Cross-cutting tasks across many files/systems | Narrow audits or checks inside a workflow |
31+
| Spawns others | ✅ Often (researchers, reviewers, fixers) | ➖ Usually spawned by a manager |
32+
| State & checkpoints | Owns plan, state, and checkpoints | Stateless or minimal state |
33+
| Typical tools | read_files, code_search, spawn_agents, run_terminal_command, end_turn | read_files, code_search, str_replace (surgical diffs) |
34+
| Output style | Plans, orchestrated diffs, verification steps | Findings, specific diffs, pass/fail signals |
35+
| Examples | Migration Coordinator, PR Shepherd, Incident Triage Conductor, Release Cut Captain | Accessibility Reviewer, API Contract Enforcer, Performance Auditor, Security Reviewer |
36+
| When to use | Complex multi-file changes needing coordination | Precise validation or localized refactors |
3037
</MarkdownTable>
3138

39+
Design checklist (use this to shape agents around task breakdown and context control):
40+
- Define the end state: what must be true when done?
41+
- List concrete steps; group them into phases (plan → edit → verify)
42+
- For each step: what files/data must be in context? how will you prove success?
43+
- Decide which steps spawn subagents (reviewer, researcher, fixer) vs. inline logic
44+
- Capture state and checkpoints so you can resume/recover deterministically
45+
- Prefer small, surgical diffs; verify before moving to the next phase
46+
3247
**Tip:** Use specialty reviewers as spawnable subagents that your context-manager agent calls at the right time in the workflow.
3348

3449
## Example: Security Coordinator Agent
@@ -154,3 +169,5 @@ With `handleSteps`, your agent can:
154169
1. Check TypeScript: `bun run typecheck` in `.agents/` directory
155170
2. Restart Codebuff to see errors
156171
3. Test with `--agent <agent-id>` to debug specific agents
172+
173+
**Tip:** Use specialty reviewers as spawnable subagents that your context-manager agent calls at the right time in the workflow.

0 commit comments

Comments
 (0)