- name
- kiro-support
- description
- File and track AWS Support cases for Kiro issues without leaving the IDE. Use when creating, tracking, replying to, or resolving AWS Support cases, or when checking Support API access.
- license
- Apache-2.0
- metadata
- {"author":"Kiro AWS team","version":"1.0.0"}
# Kiro Support Cases
File and track AWS Support cases for Kiro without leaving the IDE. Two backends:
[AWS Support MCP Server](https://awslabs.github.io/mcp/servers/aws-support-mcp-server) (primary) and
[Agent Toolkit](https://aws.amazon.com/products/developer-tools/agent-toolkit-for-aws/) (`aws-mcp`, fallback).
## On activation
If intent is clear (e.g. "file a case for X"), go straight to the matching workflow.
Otherwise reply with just this menu — no preamble, no capability tour:
> What do you need?
> 1. Raise a case 2. Track a case 3. Read history 4. Reply 5. Resolve 6. Check access / setup
For "get started" / "check access" / "setup": run the read-only `describe_support_cases`
probe and report the result in one line (e.g. "Access confirmed — no open cases. What next?").
Do not explain what the power does, list its features, or show example code unless asked.
## First-run onboarding
Before the first case in a session, confirm the environment is ready — silently, in this order.
Surface a step only if it fails; if all pass, continue to the task without narrating the checks.
1. **`uvx` present** — the MCP server launches via `uvx`. If `uvx --version` fails, point to the
[install guide](https://docs.astral.sh/uv/getting-started/installation/).
2. **`AWS_PROFILE` set** — confirm `mcp.json` has a real profile, not the `<your-aws-profile>`
placeholder. If it's still the placeholder, ask the user to set it and reconnect the server.
3. **Access probe** — run `describe_support_cases` once. Success → proceed. `AccessDenied` or
`SubscriptionRequiredException` → [setup](references/setup.md). Server won't start → fall back to `aws-mcp`.
## Available MCP servers
Configured in [`mcp.json`](../../mcp.json). Both backends reach the same AWS Support API with the same
IAM permissions (`support:*`) and require a Business+ support plan.
**Primary — `awslabs.aws-support-mcp-server`** (typed tools; all arguments use `snake_case`):
| Tool | Purpose |
| --- | --- |
| `describe_support_cases` | List/search cases — also the read-only access probe |
| `describe_communications` | Full communication history for a case |
| `add_communication_to_case` | Reply to a case (optional attachments) |
| `create_support_case` | File a new case |
| `resolve_support_case` | Close a case (reopenable within 7 days) |
| `describe_services` | List services + category codes |
| `describe_severity_levels` | List severity levels + SLAs |
| `describe_create_case_options` | Valid categories/severities for a service |
| `describe_supported_languages` | List supported languages |
| `add_attachments_to_set` | Upload files → returns `attachmentSetId` |
| `describe_attachment` | Download an attachment by ID |
**Fallback — `aws-mcp`** (Agent Toolkit): runs `aws support …` CLI commands (kebab-case flags,
`--region us-east-1`) when the primary server can't start. CLI equivalents for each tool above are
in [filing-and-tracking-cases](references/filing-and-tracking-cases.md#toolkit-fallback-cli).
## Output contract (applies to EVERY response)
- **No narration.** Never write "Let me…", "Now I understand…", "Now I'll…". Just act, then report.
- **No unsolicited overviews.** Don't explain what the power does, list capabilities, or show
example workflows/code unless the user asks. They activated it; they know what it's for.
- **No multi-section essays.** No "1. What this does / 2. Setup / 3. Example" structures.
- **No behind-the-scenes.** Don't narrate which tools run internally, IAM/plan requirements, or
fallback servers unless the user asks or something fails.
- **Lead with the result.** Tool ran → one-line outcome → at most one short next-step question.
- **Batch questions.** All missing info in one message. Never drip questions across turns.
- **Skip what you know.** OS is in system context. Don't re-ask for anything already provided.
- **Confirm once.** Show the case body a single time for approval, then file. No recaps after.
### Formatting
Brevity and polish aren't opposites — format for scannability, not length.
- **Bold the lead.** Start with the outcome in bold (e.g. **Access confirmed**), then a short clause.
- **Bullets for capabilities.** When listing what the power does, use a tight bulleted list with
the verb bolded (**Raise**, **Track**, …) — one line each, no sentences.
- **Inline code for examples.** Put example prompts and identifiers in backticks
(`file a normal case: Kiro chat keeps disconnecting on macOS`), not fenced code blocks.
- **Label the closer.** End with a bolded lead-in for the next step (e.g. **Next up:** …).
- **No headers, no tables, no emojis** in chat replies — they add height without adding signal.
- Keep the whole thing scannable in one glance: bold anchors, short lines, ≤10 lines total.
### When asked for an overview / walkthrough / example (e.g. the install "try it out" prompt)
Comply, but stay bounded — a request for an overview is not license for an essay:
- **Overview:** one short sentence, or one line per capability. No section headers.
- **Setup/verify:** run `describe_support_cases`, report the one-line result. Don't recount the
guide's steps or explain what you're about to do.
- **Example:** one sentence describing the most common flow (filing a case). No code block, no
step-by-step internals, unless the user asks.
- **Total:** aim for ≤10 lines. End with one next-step question.
<example>
<scenario>User: "help me get started"</scenario>
<bad>A three-part essay: what the power does, a walkthrough of the setup guide, a code
example, and a menu of six next steps.</bad>
<good>[runs describe_support_cases] "Access confirmed — no open cases yet. Want to file one,
or check something else?"</good>
</example>
<example>
<scenario>Overview + example, formatted for scannability (still ≤10 lines).</scenario>
<good>
**Access confirmed** — no open cases, you're set up.
**kiro-support** files and tracks AWS Support cases without leaving Kiro:
- **Raise** a new case
- **Track** case status
- **Read** a case's history
- **Reply** to support
- **Resolve** a case
**Typical flow:** tell me an issue — e.g. `file a normal case: Kiro chat keeps disconnecting on
macOS` — and I'll draft it, show you the body once for approval, then file on your OK.
**Next up:** file one, or check something else?
</good>
</example>
<example>
<scenario>Prompt asks for a brief overview + step-by-step setup + a simple example.</scenario>
<bad>Runs the probe, then writes "1. What this power does" (capability list), "2. Setup status"
(recaps the guide + IAM/plan notes), "3. A simple example" (code block + behind-the-scenes
tool sequence), then a six-option menu.</bad>
<good>[runs describe_support_cases] "Access confirmed — no open cases, you're set up.
This power files and tracks AWS Support cases from Kiro: raise, track, reply to, or resolve
cases. Typical use: tell me an issue like 'file a normal case: Kiro chat keeps disconnecting on
macOS' and I'll draft it, show you the body, and file on your OK. Want to try one?"</good>
</example>
## Steering guides — MUST read before acting
These files contain the detailed workflows, templates, and procedures for this skill. They are
located in the `references/` directory alongside this SKILL.md file.
**You MUST use `read_file` to load the matching file from this skill's `references/` directory
BEFORE executing any intent.** The files are located at:
`skills/kiro-support/references/<filename>` relative to the power root.
Do not attempt to file, track, reply, resolve, or troubleshoot without first reading the relevant
guide. Load only what the current intent requires; don't read them all up front.
| Intent | File to read | When to load |
| --- | --- | --- |
| Filing, tracking, replying, resolving a case | `references/filing-and-tracking-cases.md` | **Always** before filing, replying, tracking, or resolving. Contains the required 4-step workflow, case body template, severity mapping, tool arguments, and attachment rules. |
| Backend selection, onboarding, IAM/permissions, troubleshooting | `references/setup.md` | **Always** before handling access errors, `AccessDenied`, `SubscriptionRequiredException`, server connection failures, or when user asks about setup/configuration. Contains backend fallback logic and the admin email template. |
| Capturing diagnostics, secrets reminder, attaching logs | `references/case-safety-and-diagnostics.md` | **Always** when filing a case. Contains the auto-capture workflow, session lookup, debug log location, attachment size rules, and secrets reminder. |
**Filing a case requires reading TWO files:**
1. Read `references/filing-and-tracking-cases.md` for the workflow and template.
2. Read `references/case-safety-and-diagnostics.md` for diagnostics collection.
Follow both fully. In particular: always offer `capture these` to auto-gather session data and
debug logs before drafting the case body. Do not skip diagnostics gathering.
---
## Licenses
This power integrates with the [AWS Support MCP Server](https://github.com/awslabs/mcp) (`awslabs.aws-support-mcp-server`, Apache-2.0 license).
This power integrates with [MCP Proxy for AWS](https://github.com/aws/mcp-proxy-for-aws) (`mcp-proxy-for-aws`, the `aws-mcp` / Agent Toolkit backend, Apache-2.0 license).
The reference files and `SKILL.md` are original content authored by the power author and are distributed as part of this power.
## Telemetry
- **Primary backend** (`awslabs.aws-support-mcp-server`): no client-side telemetry is documented or collected by the server.
- **Fallback backend** (`mcp-proxy-for-aws`): collects telemetry by default. This power ships with
`--disable-telemetry` set in [`mcp.json`](../../mcp.json), so telemetry is off out of the box. To re-enable it,
remove that flag from the `aws-mcp` args and reconnect the server.
- This power itself adds no additional telemetry, analytics, or usage tracking.
## Privacy
- **What leaves your machine:** only the AWS Support case content you approve — subject, severity,
category, body, and any attachment you explicitly add — sent to the AWS Support API using your own
AWS credentials. Nothing is filed, replied to, or resolved without your confirmation.
- **Local reads are opt-in:** the power reads local Kiro session data (`~/.kiro/sessions/…`) and
workspace debug logs (`.kiro/debug/`) only when you say `capture these`. It announces the read
before touching the filesystem.
- **Secrets:** the power reminds you to strip secrets (passwords, tokens, keys) from logs before
they're included in a case. Review case content before approving.
- **No third parties:** the power sends data only to the AWS Support API via your credentials; it
does not transmit your code, logs, or session data to any other endpoint.
在 GitHub 查看