aws-samples / aws-samples/sample-autonomous-cloud-coding-agents

feat(registry): capability descriptors for dynamic environment config

Open
#481 1 comment 0 reactions 0 assignees View on GitHub
agent-runtime enhancement registry
Dominant language
TypeScript
Stars
143
Forks
46
Avg merge
3d 9h
Merged PRs (30d)
20

Description

**Context:** ROADMAP.md → Capability descriptors

---

## Component

Agent (Python runtime)

## Describe the feature

Structured **metadata per registry asset** declaring:

**Provides:** tool surface, permissions required, resource limits, Cedar actions introduced.

**Requires:** runtime dependencies, network egress domains, minimum compute profile.

Agent runtime uses descriptors to configure the execution environment dynamically—MCP servers, context injection, sandbox permissions—rather than static blueprint lists. Registry is evaluated by Change Manifest **L1** (`tool_needed` validation) and **L5** (knowledge resolution).

## Use case

Static blueprint lists do not scale as asset types grow. Runtime needs machine-readable requirements to provision MCP servers, context injection, and sandbox permissions.

## Proposed solution

1. JSON Schema for `CapabilityDescriptor` stored alongside asset artifact.
2. Resolver validates descriptor at task start; fails closed on mismatch with compute profile.
3. Gateway + Cedar auto-generate allowlists from descriptor actions.
4. Validator in registry publish path.
5. Golden fixtures in `contracts/` or `cdk/test/`.

## Other information

- Child of #246; pairs with extended tool capability tiers draft.
- Design context: `docs/design/COMPUTE.md`.

- [ ] This might be a breaking change

Contributor guide

Open the contributing guide

Research direction

Start with ROADMAP.md and docs/design/COMPUTE.md, then inspect the registry publish path and the contracts/ or cdk/test/ locations mentioned for golden fixtures. The proposed work spans the descriptor schema, task-start resolver, Gateway and Cedar allowlists, publish validation, and fixtures; done means these pieces support validated dynamic environment configuration.

Written by the indexing model from the issue text.

Assessment

Tech stack
json, python
Domain
backend, devops, security
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.