Agent Templates

Kida ships six built-in AMP message types, each backed by a primary agent-report template for rendering structured AI output as formatted GitHub comments, step summaries, releases, and terminal output. Some message types also have multiple surface-specific template variants. Each template consumes structured JSON from an AI agent and produces surface-appropriate output with severity badges, collapsible sections, diff suggestions, and more.

Template catalog

All schemas are inschemas/amp/v1/. All templates are in templates/.

Code review

The code-review template renders AI review findings as a structured PR comment with severity grouping, clickable file permalinks, and diff suggestions.

Schema

Finding fields:

Example input

{
  "agent": "Claude",
  "confidence": 0.87,
  "summary": "Found 2 issues: 1 security warning, 1 bug.",
  "repository": "myorg/myrepo",
  "sha": "abc1234",
  "min_confidence": 0.5,
  "findings": [
    {
      "file": "src/db.py",
      "line": 42,
      "severity": "warn",
      "category": "security",
      "title": "SQL injection risk",
      "message": "Query built with f-string interpolation.",
      "diff": "- query = f\"SELECT * FROM users WHERE id = {user_id}\"\n+ query = \"SELECT * FROM users WHERE id = %s\"",
      "confidence": 0.94
    },
    {
      "file": "src/core.py",
      "line": 156,
      "severity": "warn",
      "category": "bug",
      "title": "Unhandled None return",
      "message": "get_template() can return None but caller assumes Template.",
      "confidence": 0.82
    }
  ],
  "stats": {
    "files_reviewed": 8,
    "findings_count": 2,
    "by_severity": { "error": 0, "warn": 2, "info": 0, "suggestion": 0 }
  }
}

Rendered output

The template produces:

• A header with the agent name and overall confidence badge
• A summary line
• Findings grouped by severity (error first, then warn, info, suggestion)
• Each finding shows a severity badge, file permalink (owner/repo/blob/sha/file#L42), and the message
• Findings with adiff field render as a fenced code block with diffsyntax highlighting
• Findings belowmin_confidence are collapsed inside a <details>block
• A stats footer with files reviewed and severity breakdown

PR summary

Renders an auto-generated PR description with what changed, why, risk areas, and review suggestions.

Schema

Example input

{
  "agent": "Claude",
  "title": "Add rate limiting to API endpoints",
  "what_changed": "Added token-bucket rate limiting to all `/api/v2/` endpoints.",
  "why": "Production is seeing 10x traffic spikes from a single client.",
  "change_size": "medium",
  "files": [
    { "path": "src/middleware/rate_limit.py", "change_type": "added", "summary": "New rate limiter middleware" },
    { "path": "src/api/router.py", "change_type": "modified", "summary": "Wired rate limiter into router" }
  ],
  "risk_areas": [
    { "area": "Redis connection pool", "reason": "Rate limiter uses Redis; pool exhaustion could block requests" }
  ],
  "breaking_changes": [],
  "test_coverage": "Unit tests for token bucket logic. Integration test for rate-limited endpoint.",
  "suggested_labels": ["enhancement", "api"],
  "suggested_reviewers": ["alice"]
}

Rendered output

• Breaking changes alert (if any) with a> [!CAUTION]block
• What/why sections
• File change table with change type badges
• Risk areas with explanations
• Test coverage summary
• Suggested labels and reviewers
• Diff stats footer

Deploy preview

Renders deploy status with a preview URL, bundle size deltas, and Lighthouse performance scores.

Schema

Rendered output

• Status badge (pass/fail/building)
• Clickable preview URL
• Bundle size table with per-entry deltas
• Lighthouse score badges (color-coded: green >= 90, yellow >= 50, red < 50)
• Per-page status table

Dependency review

Renders dependency change analysis with vulnerability detection and license compliance.

Schema

Rendered output

• Overall risk badge
• Added/updated/removed dependency tables
• Vulnerability alerts with severity badges, CVE links, and fix versions
• License compliance warnings

Security scan

Renders security findings grouped by severity with CWE/CVE references and remediation steps.

Schema

Finding fields:

Rendered output

• Overall status badge and policy compliance
• Findings grouped by severity (critical first)
• Each finding shows severity badge, CWE/CVE links, file permalink
• Remediation steps and diff suggestions
• Stats footer with severity breakdown

Release notes

Renders categorized release notes with contributor highlights, breaking changes, and dependency digests. This type has four template variants for different surfaces.

Schema

Template variants

Rendered output

• Version header with release date and compare link
• PRs grouped by label category (features, fixes, docs, etc.)
• Breaking changes callout
• Contributor list with first-time contributor highlights
• Dependency update digest (for dependabot/renovate PRs)
• Diff stats and benchmark deltas

Customizing templates

Override a built-in template

Copy the built-in template and modify it:

cp templates/code-review-report.md .github/templates/my-code-review.md

Then point your workflow at the custom template:

kida render .github/templates/my-code-review.md --data review.json --mode markdown

Or in the GitHub Action:

- uses: lbliii/kida@v0.3.3
  with:
    template: .github/templates/my-code-review.md
    data: .amp/collected/code-review.json

Template patterns

Built-in agent templates follow consistent patterns you can reuse:

Variable defaults at the top:

{% set min_confidence = min_confidence | default(0.5) %}
{% set repository = repository | default("") %}
{% set sha = sha | default("") %}

Severity badges:

{{ finding.severity | badge }}

Conditional file permalinks:

{% if repository and sha %}
[`{{ finding.file }}#L{{ finding.line }}`](https://github.com/{{ repository }}/blob/{{ sha }}/{{ finding.file }}#L{{ finding.line }})
{% else %}
`{{ finding.file }}:{{ finding.line }}`
{% endif %}

Collapsible low-confidence findings:

{% if finding.confidence < min_confidence %}
<details>
<summary>{{ finding.title }} (low confidence: {{ (finding.confidence * 100) | round | int }}%)</summary>

{{ finding.message }}

</details>
{% endif %}

Available filters

Agent templates use these Kida filters:

Next steps

• Agent Message Protocol — the protocol spec, surfaces, and composition modes
• Agent Integration Tutorial — end-to-end setup from agent config to PR comments
• GitHub Action — CI report rendering with the Kida action