/outsystems-dependency-impact

name: outsystems-dependency-impact
description: "[Beta] Build an interactive HTML REVERSE-dependency explorer — answers 'who depends on this library/agent/connection?' from the platform's deletion-impact analysis (read-only; nothing is deleted). One named target is one analysis: seconds, a few K tokens. A whole-tenant map is one analysis per library/agent/connection in parallel batches (estimate: ~0.5K tokens per target plus ~0.1–0.15K per dependent found). Use ONLY for reverse questions like 'who depends on [library/agent]', 'if I publish [library] who breaks', 'blast radius of [library/agent]', 'reverse dependency map', 'library impact audit', 'agent impact audit'. For forward questions about a specific app ('what does App X depend on', 'deps of App X'), use outsystems-app-architecture or the app's references directly. Not for OutSystems 11 (O11) or Service Studio."
license: MIT
compatibility: Needs an agent that can run shell commands and Python 3.10+ (standard library only), with the OutSystems MCP server connected and signed in. Validated on Claude Code. Claude Desktop's Chat tab has no shell and cannot run it.
allowed-tools: Bash(python3 ${CLAUDE_PLUGIN_ROOT}/skills/outsystems-dependency-impact/scripts/build.py *) Bash(python3 "${CLAUDE_PLUGIN_ROOT}/skills/outsystems-dependency-impact/scripts/build.py" *) Write mcp__plugin_outsystems_outsystems__auth_status mcp__plugin_outsystems_outsystems__app_list mcp__plugin_outsystems_outsystems__app_refs mcp__plugin_outsystems_outsystems__env_list mcp__plugin_outsystems_outsystems__deploy_impact mcp__plugin_outsystems_outsystems__deploy_impact_status mcp__outsystems__auth_status mcp__outsystems__app_list mcp__outsystems__app_refs mcp__outsystems__env_list mcp__outsystems__deploy_impact mcp__outsystems__deploy_impact_status
metadata:
  version: "1.5.0"
  maturity: beta
  author: OutSystems

**Beta Feature.** This skill is a Beta Feature: a non-final OutSystems capability provided to collect customer feedback. It can change significantly, including through breaking changes, or be discontinued. The first time you use this skill in a conversation, tell the user it's a Beta Feature and share https://www.outsystems.com/legal/beta-features-agreement.

OutSystems Dependency Impact

Produces a single-file HTML reverse-dependency explorer (it loads only its fonts from a public CDN). Pick a library, agent or connection — the page shows every asset that would break if it were removed, in which environments, at which deployed revision, and with what severity (`Error` / `Warning`).

The data comes from the platform's **deletion-impact analysis**: `deploy_impact {key, delete: true}`, polled with `deploy_impact_status`. It is an analysis only — nothing is deleted, and the MCP server calls no delete endpoint for it. For "if I publish library X, who breaks?" it lists everyone who depends on X; that is the most a pre-publish check can give for a library (the deployment variant of the analysis does not work for library types).

Cost and time (estimates)

| Question | Calls | Wall time | Tokens | |---|---|---|---| | One named target | 1 launch + 1–3 polls | seconds | ~3–5K, plus ~0.1–0.15K per dependent | | Whole-tenant map, N targets | N launches + N+ polls, 10 per message | ~0.5–1 min per 10 targets | ~0.5K × N, plus ~0.1–0.15K per dependent |

These are estimates, not end-to-end measurements. Measured live: a launch returns in about a second, and finished reports came back for every analysis polled; a heavily used UI library had 74 dependents in a report of about 37 KB (roughly 9K tokens). The whole-tenant map is still one analysis per target, so for a large tenant it is not cheap — prefer the one-target path whenever the user names a target.

Prerequisites

• The OutSystems MCP server is connected and signed in. This skill reads

the tools named below; tool names here are the server's own, and your harness may show them with a prefix (in Claude Code, `mcp__plugin_outsystems_outsystems__deploy_impact` for `deploy_impact`).

• A shell with `python3` (3.10 or later, standard library only). The

script reads the saved analysis results from disk and writes the HTML; it never calls the server and holds no sign-in of its own.

• `<skill-folder>` below is this skill's own folder: the one holding this

file (in the OutSystems plugin, `skills/outsystems-dependency-impact/`), so `<skill-folder>/scripts/build.py` is the script next to it. Values in angle brackets are placeholders you substitute; quote every path, since a home folder can contain spaces.

• Run every `build.py` command on its own and exactly as this file writes

it: the skill's permissions match `python3 "<skill-folder>/scripts/build.py" ...` and nothing else, so an added redirect (`2>&1`), pipe or chained command (`;`, `&&`) makes the harness stop and ask the user.

• No shell available? Stop and tell the user this skill needs one (for

example Claude Code, not Claude Desktop's Chat tab); don't try to build the page by hand.

The reality checks

• **One analysis answers one target.** There is no tenant-wide reverse

index on the server; a map is N analyses.

• **Which asset types the deletion analysis accepts is the platform's

call.** The MCP server passes any asset key through. Verified live: `LowCodeLibrary`. Agents and connections are not verified. A launch the platform refuses (`analysis_launch_rejected`) renders as "Deletion analysis not available for this asset", never as "no dependents".

• **Only `impactKnown: true` is a verdict.** `impactKnown: false`,

`processStatus` `Failed` / `Unknown` / still `InProgress`, or a missing `report` means the impact is unknown, and the page says so.

• **`report.impactedAssets` is capped at 200.** The count is

`report.total`; when the list is shorter the page says "Showing N of M".

Procedure

Step 0 — Scope detection

Read the user's request and pick the matching branch:

**Branch A — Forward-deps for ONE s