Bind this session to a project in the rawgentic workspace, or manage project activation. Use when starting work on a specific project, when the system says "Multiple projects active", or when you want to enable/disable projects. Invoke with /rawgentic:switch followed by a project name, "off <name>" to deactivate, or no args to list all.
Bind this session to a project in the rawgentic workspace, or manage project activation. Use when starting work on a specific project, when the system says "Multiple projects active", or when you want to enable/disable projects. Invoke with /rawgentic:switch followed by a project name, "off <name>" to deactivate, or no args to list all.
argument-hint
project name (e.g., my-app), "off <name>" to deactivate, or empty to list
You are the rawgentic project switcher. Your job is to bind the current session to a project so all subsequent rawgentic workflow skills and hooks operate on the correct project. You do NOT deactivate other projects — multiple projects can be active simultaneously for concurrent sessions.
Switch / Bind Project — /rawgentic:switch
Run through the steps below sequentially.
Step 1: Parse Input
The user provides a name, path, or subcommand. They might also use natural language.
No argument and no project mentioned → Go to Step 2 (list mode).
"off " → Go to Step 6 (deactivate mode).
Name or path provided → Use it as the search target in Step 3.
Step 2: List Mode
Read .rawgentic_workspace.json from the primary working directory.
File missing → STOP. Tell the user: "No rawgentic workspace found. Run /rawgentic:new-project first."
Then ask: "Which project do you want to bind this session to?"
Step 3: Find in Workspace
Read .rawgentic_workspace.json from the primary working directory.
File missing → STOP. Tell the user: "No rawgentic workspace found. Run /rawgentic:new-project first."
Malformed JSON → STOP. Tell the user: "Workspace file is corrupted."
Search the projects array:
Match by name first (case-insensitive comparison).
Then match by path (normalize both paths).
If not found: List all projects and suggest: "No project matching ''. Did you mean one of these?"
Step 4: Verify Directory Exists
Check that the target project's path directory exists on disk.
Path resolution: Resolve relative paths against the workspace root directory.
If missing: Warn: "The directory <path> no longer exists. Run /rawgentic:new-project <name> to re-create, or /rawgentic:switch off <name> to deregister."
Step 5: Bind Session
Read .rawgentic_workspace.json, then:
Set the target project's active to true (if it wasn't already — this enables a project that was previously deactivated).
Update the target's lastUsed to the current ISO 8601 timestamp.
Write the updated workspace file back (full read-modify-write).
Do NOT set any other project's active to false. Multiple projects can be active simultaneously.
Register in session registry: Append a line to claude_docs/session_registry.jsonl:
{"session_id":"<your session_id>","project":"<target project name>","project_path":"<target project path>","started":"<current ISO 8601 timestamp>","cwd":"<workspace root>"}
Create the file and claude_docs/session_notes/ directory if they don't exist.
How to get your session ID — use the per-session env var, NOT the shared file:
The session ID is in $CLAUDE_CODE_SESSION_ID, set per Claude Code process (so it is unique and correct even when multiple sessions run concurrently). Register in two expansion-free Bash calls so the command contains no $(...) command substitution: a command containing $(...) (or backticks) is flagged "Contains expansion" and always triggers a permission prompt that no permissions.allow rule can suppress — keeping it expansion-free lets a Bash(printf:*) / Bash(date:*) / Bash(printenv:*) allow rule auto-approve the bind.
Call 1 — read the session ID and timestamp (two allowlistable leading binaries, no command substitution):
printenv CLAUDE_CODE_SESSION_ID; date -u +%Y-%m-%dT%H:%M:%SZ
Call 2 — append the registry line, inlining those two values as literals (starts with printf, no $(...), only >> redirection, so it matches a Bash(printf:*) allow rule):
printf'{"session_id":"%s","project":"%s","project_path":"%s","started":"%s","cwd":"%s"}\n'"<SESSION_ID from call 1>""<target project name>""<target project path>""<TIMESTAMP from call 1>""<workspace root>" >> claude_docs/session_registry.jsonl
Because $CLAUDE_CODE_SESSION_ID is per-process, reading it in call 1 and writing in call 2 is still race-free — there is no shared state between the two calls to corrupt.
Do NOT read claude_docs/.current_session_id as the source: that file is shared across all sessions and is overwritten by every session on every prompt, so under concurrent sessions it can return another session's id and bind the wrong session. If printenv prints nothing (only older Claude Code that does not set the env var), STOP and ask the user rather than guessing. (The legacy name $CLAUDE_SESSION_ID is not set — the correct variable is $CLAUDE_CODE_SESSION_ID.) Do NOT invent a session ID.
Report:
Bound to: <name> (<path>)
Configured: yes/no
If configured is false: Suggest: "This project hasn't been configured yet. Run /rawgentic:setup." Then skip Step 5b entirely — there is no config to check for staleness.
If configured is true: Proceed to Step 5b before confirming "Ready."
Step 5b: Config Staleness Check
After binding and before the "Ready" confirmation, run these checks in order:
1. Workspace-level: defaultProtectionLevel
Read .rawgentic_workspace.json. If the top-level field defaultProtectionLevel is missing:
Prompt the user:
Your workspace is missing a default protection level.
Choose a workspace-wide default for new or unconfigured projects:
- sandbox — No guards active. Good for POC / playground projects.
- standard — Blocks destroy + mutate ops on production, 6 common security patterns.
- strict — All guards active. Full production projects.
Which level? (sandbox / standard / strict)
Wait for the user's choice. Validate it is one of sandbox, standard, strict.
Read .rawgentic_workspace.json, add "defaultProtectionLevel": "<choice>" at the top level, and write it back (full read-modify-write).
Confirm: "Set workspace defaultProtectionLevel to ."
This prompt runs once — subsequent binds see the field and skip.
2. Project-level: universal field check
Check the project's .rawgentic.json for the following universal fields (hardcoded list):
version
project
repo
protectionLevel
custom
This list is intentionally small — it includes only fields that every project should have regardless of type. Optional sections (testing, database, services, infrastructure, deploy, security, ci, formatting, documentation) are NOT checked because projects may legitimately omit them.
Compare field presence only — do not validate values or nested structure.
If any universal fields are missing: Print an advisory warning:
Config advisory: your .rawgentic.json is missing: <comma-separated list of missing fields>
Run `/rawgentic:setup` to update your config (existing values will be preserved).
If no fields are missing: Silent pass — print nothing.
2b. Feature-gap staleness nudge (#234)
The universal-field check above only catches a malformed/old-shape config. It does
NOT catch a project that predates newer setup-requiring features (adversarial
review, model routing, peer consult, design artifact, …) — that project has a valid
config, it's just behind. hooks/post_update_reconcile.py's SessionStart pass nudges
these once per plugin version, but switching to a project is the moment to surface
its own gap. Run the per-project staleness check for the project you just bound
(<name> = the target project's name; resolve the workspace + claude_docs paths as
in Step 5):
It prints an advisory line (or nothing) — surface any output to the user verbatim.
This is advisory, never blocking: it always shows the gap on an explicit switch
(no once-per-version gate, unlike the SessionStart --staleness-active pass), and it
respects the workspace-level "setupPrompt": false opt-out. Fail-open: a non-zero
exit or empty output means "nothing to nudge" — continue.
3. Headless Access Check
If the current session has RAWGENTIC_HEADLESS=1 set (headless mode), check the target project's headlessEnabled field in .rawgentic_workspace.json. The field accepts a bool (legacy) or an object {"enabled": bool, "triggers": [...], "auth": "..."} (#165) — apply the SAME verdict the session-start gate computes:
If headlessEnabled is true: Silent pass — headless mode allowed, any trigger.
If it is an object with enabled: true: allowed only when triggers is absent, OR $RAWGENTIC_HEADLESS_TRIGGER is a member of the triggers array. A non-member, an unset trigger env, or a malformed triggers value fails CLOSED — STOP and tell user: "Headless mode for [project-name] does not allow this trigger (RAWGENTIC_HEADLESS_TRIGGER is not in the headlessEnabled.triggers allowlist)."
If headlessEnabled is false, {"enabled": false, ...}, missing, or any other shape: STOP and tell user:
"Headless mode is not enabled for [project-name]. Run /rawgentic:setup to enable it, or set headlessEnabled: true in the project's entry in .rawgentic_workspace.json."
If not in headless mode: skip this check entirely.
4. Confirm Ready
After all checks complete, print the final confirmation:
"Ready. All rawgentic workflow skills will use <path>/.rawgentic.json for this session."
Step 6: Deactivate Mode (/rawgentic:switch off <name>)
Find the project in the workspace (same as Step 3).
Check claude_docs/session_registry.jsonl for sessions bound to this project in the last 24 hours.
If recent sessions found → warn: "There are recent sessions bound to . Deactivating will not unbind them, but new sessions won't auto-bind to it. Continue?"
Set active: false for the project in .rawgentic_workspace.json.
Write the updated workspace file.
Report: "Deactivated . It won't appear as an option for new sessions until reactivated."
