Install SDD: next slice of per-file specs

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSM4FVUbnP4PJJ9sawMfLb

Author: claude

packages/the-framework/dashboard/lib/quota.SPEC.md

@@ -0,0 +1,7 @@
+Feeds the dashboard's quota panel: where the account's quota stands against its limits, and what Auto PM last decided. Both are asked of the daemon every thirty seconds while the panel is open, since the daemon already keeps a cached reading and answering is cheap — unlike the driver read behind it — and since Auto PM runs on the daemon's clock, so the panel only learns of a new decision by asking again.
+
+Neither answer exists until the first reply arrives, and a host that runs no Auto PM has no decision to report. A failed request keeps the last known reading rather than blanking it, because an empty bar would read as "nothing used".
+
+## Before modifying/creating SPEC.md files
+
+You must always read and respect https://raw.githubusercontent.com/brillout/sdd/refs/heads/main/sdd.md

packages/the-framework/dashboard/lib/quota.test.SPEC.md

@@ -0,0 +1,5 @@
+What the tests cover: the quota reading the dashboard shows. There is no reading at all until the first answer arrives — distinct from "nothing used" — after which the daemon's answer is reported as-is. The reading refreshes on its thirty-second interval. A failed refresh keeps the last reading rather than blanking it, since an empty bar would read as nothing used. Closing the panel stops the refreshing.
+
+## Before modifying/creating SPEC.md files
+
+You must always read and respect https://raw.githubusercontent.com/brillout/sdd/refs/heads/main/sdd.md

packages/the-framework/dashboard/lib/remote-target.SPEC.md

@@ -0,0 +1,7 @@
+Remembers which saved device the user has picked to run the next agent on, so the Start form can show the choice and the submit can attach it as the agent's relay target. No device picked means the agent runs on one of the ordinary run targets — this machine or a GitHub Actions runner.
+
+The choice lasts only as long as the page: it is never stored, and in particular never becomes a preference, because a device's token is a per-browser secret and where one agent is sent is a transient decision rather than a saved setting.
+
+## Before modifying/creating SPEC.md files
+
+You must always read and respect https://raw.githubusercontent.com/brillout/sdd/refs/heads/main/sdd.md

packages/the-framework/dashboard/lib/resume-command.SPEC.md

@@ -0,0 +1,23 @@
+Builds the shell one-liner that reopens an agent's driver session in a terminal, so a conversation can be continued outside the dashboard.
+
+## Business logic
+
+### Picking a session back up outside the dashboard
+
+#### User story
+
+The user wants to carry on an agent's conversation in their own terminal rather than in the dashboard. The dashboard knows the session id but the conversation was previously unreachable: there was no way to name it.
+
+#### Business logic
+
+The command recreates the directory the agent ran in, changes into it, and resumes the session by its id. With no directory recorded — an older agent, or one that never opened a session — the session id alone is handed over instead, so the caller can say which of the two it gave the user. An agent with no session id at all yields nothing to copy.
+
+#### Rationale
+
+The session id alone is not enough: the coding-agent CLI finds a session by the directory it ran in, and that directory is usually gone by the time the user wants it, because an agent that finishes cleanly has its worktree removed. An empty directory at the original path is enough for the CLI to match the session and read the conversation back.
+
+The command deliberately presets no permission mode. It lands in someone's own terminal, and what the reopened agent is allowed to do is that person's call to make at their prompt.
+
+## Before modifying/creating SPEC.md files
+
+You must always read and respect https://raw.githubusercontent.com/brillout/sdd/refs/heads/main/sdd.md