env-check

Env check — Cloud agent vs local workspace

When to use

Run early in a session (or when continuity / credentials / browser tests / local CLIs matter) to classify the host. Use the result as supporting context. The Hub continuity fallback is triggered when ADHD Hub MCP is missing, errored, unauthorized, or otherwise unavailable — not merely because the runtime looks like a cloud sandbox.

Quick detect

Resolve scripts/check_runtime.sh relative to this skill’s install location (not the ADHD Hub source tree alone).

Supported layouts:

Layout	Example command
ADHD Hub source checkout	`bash skills/env-check/scripts/check_runtime.sh`
Project-scoped install (`.agents/skills`)	`bash .agents/skills/env-check/scripts/check_runtime.sh`
Global / plugin install	From the installed `env-check` skill directory: `bash scripts/check_runtime.sh`

# Project-scoped (common after adhd-hub sync-project / setup --project-skills):
bash .agents/skills/env-check/scripts/check_runtime.sh

Output lines:

RUNTIME_ENV: CLOUD_AGENT or RUNTIME_ENV: LOCAL_WORKSPACE
RUNTIME_SIGNALS: … (container / Cursor markers; informational)
HUB_CONTINUITY_TRIGGER: MCP unavailable …
LOCAL_SKILL_CLIS: … (CLOUD_AGENT: do not assume; LOCAL_WORKSPACE: may be available)

Detection quality

The script does not treat $USER=root as cloud by itself (false positives on local root shells). It combines:

Cursor Cloud markers: CURSOR_AGENT, __CURSOR_SANDBOX_ENV_RESTORE, /tmp/cursor/cloud-agent-transcripts, /exec-daemon, plus agent sockets when paired with container cues
Container cues (/.dockerenv, cgroup) as supporting signals only — local Docker alone stays LOCAL_WORKSPACE

Do not invent “I'm in cloud” from a single weak hint. Prefer the script output when unsure.

CLOUD_AGENT guidelines

When RUNTIME_ENV: CLOUD_AGENT (Cursor Cloud / remote sandbox):

Do not assume machine-installed local skill binaries or CLIs exist (examples: graphify, other opt-in companions installed on a developer laptop). Cloud/remote sandboxes often cannot see those tools even when project rules mention them.
If a required local CLI is missing: one-line notice, then continue via repo tools (Read/Grep/Glob/tests) and any committed artifacts such as graphify-out/ when present. Never fabricate graphify graph output or Hub continuity/progress/thread state.
No native browser GUI or macOS/Windows desktop binaries; prefer headless tests and CLI verification
Prefer injected environment / OIDC / platform secrets over committing or relying on .env.local
Local-only assumptions (opening localhost GUIs, host Docker socket from the agent VM, Tailscale to a private Hub) may fail — plan accordingly
If Hub MCP tools are missing/errored/unauthorized: on the first substantial Hub-worthy turn, the first line of the reply MUST say Hub MCP is not available, plus a short fix hint (MCP URL → this Hub's /mcp, ADHD_HUB_AUTH_TOKEN, restart the agent; skip/cancel Auth if it hangs until Hub OAuth is enabled). Never invent Hub state or claim a Hub write succeeded
Fallback still applies when Hub MCP is down: when issue-write access exists and the identity is on Hub Inbox authors, open/update a forge/GitHub issue titled [ADHD] … with Goal / Focus / Next / Resume (or Now / Done / Next / Return). Optional labels adhd-hub, project:<slug>, source:cursor — skip labels if the token cannot set them. Recommended: append Made with [ADHD Progress Hub](https://github.com/uniskela/adhd-hub) under a non-imported heading (e.g. ## Attribution). No secrets, private Hub URLs, internal hosts/IPs, absolute local paths, or transcripts. Otherwise state that continuity persistence is unavailable and continue authorized work.

LOCAL_WORKSPACE guidelines

When RUNTIME_ENV: LOCAL_WORKSPACE (including local Docker / localhost):

Machine-installed local skill CLIs (e.g. graphify) may be available — use them when present; still do not invent output if a binary is missing
Local Docker, localhost services, and host browsers are OK when the operator's machine provides them
Prefer Hub MCP (resolve_project, session_digest, upsert_progress, …) when it is reachable
Same no-hallucination rule if MCP is down: acknowledge + fix hint; never invent Hub state; use the forge [ADHD] mailbox only when write access + Inbox authors allow it

Continuity priority

MCP unavailable? → loud first-line warning + forge [ADHD] fallback (if allowed) — this is the Hub continuity trigger (still true on CLOUD_AGENT)
Runtime env → choose tests, credentials, tooling constraints, and whether local skill CLIs are expected
Prefer Hub MCP whenever it is available, on either runtime

See also: adhd-hub-session (full Hub protocol), docs/forge-issue-inbox.md.