[cosmosquark]

Agents should identify what the selected game needs and whether the user wants a new installation or an action on an existing server. This PR adds that assessment while limiting hard approval gates to new software installation.

• Add alphagsm-server-requirements (36 skills total), with source-backed requirements for the exact canonical game/version: platform, manager/install tools, runtime, provider/assets, settings, ports and resources. Inspect actual files and aliases, not just registered tags.
• For a vague run request with no matching game installed, ask to install it. If the game already exists and the action/target is unspecified, clarify whether to start it, change it or create another. Explicit actions on identified existing servers and explicit/approved installs proceed without another confirmation. Assessment-only requests stay read-only, including manager state: no implicit manager bootstrap/recreation or image acquisition to make a diagnostic available.
• AlphaGSM code, supported helpers, repository image pulls/builds, project-local pinned dependencies and managed mod/plugin prerequisites are part of the requested work and have no extra approval blockers. An approved game installation includes its managed recipe; SourceMod includes supported MetaMod setup. Existing Docker has no first-use consent gate.
• Ask before genuinely new external software outside that managed scope, including a missing host Docker engine/Compose, host runtime or system tool. Reuse approval within its installation scope and continue independent work while a reply is pending.
• Prefer the repository's Docker manager/game runtimes and AlphaGSM images published by or built from SectorAlpha/AlphaGSM. Accurately distinguish the supported host-socket manager from true Docker-in-Docker. Update entry/specialist routing, guides, discovery, changelog and evaluation scenarios.

Compatibility with companion PR #39

Reviewed against PR #39 commit 901e32a5; this PR remains independently based on master and does not include #39's implementation.

• Detect actual installed capabilities. Shared anonymous Workshop staging is separate from native item/collection activation; older checkouts retain their supported command paths.
• Add conditional tModLoader release/startup/world guidance, native content ownership/cleanup and process-versus-Docker load evidence. Account for Squad RCON and Wreckfest's limited Linux UDP health without invented provider auth or game metadata.
• Pending native Workshop validation stays pending; stock server support and downloaded/configured files do not establish engine activation.
• Both merge-order simulations have one conflict: changelog.txt. Resolve it by combining the October 14 entries under one date, preserving Add anonymous Workshop downloads and native server mod activation #39's October 13 entries and Add game requirements and scope new-software installation approvals #38's October 12 entries. DEVELOPERS.md and all implementation files merge automatically. The second PR to land should rebase onto updated master with this release-note resolution.
• Structural validation passed on Add game requirements and scope new-software installation approvals #38 alone and a temporary combined snapshot with the changelog resolved: 36 skills and 24 validator tests in each. Inspection/forward review found no remaining actionable compatibility gap. No Add anonymous Workshop downloads and native server mod activation #39 game lifecycle or native-load claims are made by these checks.

Validation:

• make lint: 10.00/10.
• make skills-check: all 36 skills valid.
• Focused validator tests: 24 passed; bundled skill-creator validation passed for the new skill.
• Read-only baseline/forward scenarios and independent review cover absent/existing/multiple/explicit-action requests, aliases, assessment-only requests, managed image/helpers/content, and external host software. Findings were corrected.
• git diff --check passed.
• October 14 review: corrected the read-only manager diagnostic edge and rechecked assessment-only versus explicit-start scenarios; no remaining actionable findings.
• Cloud CI on the preceding head encountered a Vintage Story smoke metadata request failure (Network is unreachable); the official endpoint now responds HTTP 200. The failed smoke batch was requeued; the compatibility update will run normal checks. This branch is now current with master, including Add optional setup tour and Sector Alpha agent with guided binding #40 and Retry transient SteamCMD failures and add server framework packs #43. The earlier summarize-tests failure came from leftover integration-recheck artifacts on a rerun of run 39076108599. Checks are running again on this update.

Scope: agent instructions and documentation, not a new CLI consent mechanism. No game/runtime code, dependency manifest or test implementation changes; no new software was installed during this work. Live smoke/integration/Docker lifecycles were not rerun for this documentation change. Scenario evaluations are inspection exercises, not live lifecycle claims.