Claude Code slash commands: a full inventory (source investigation notes)

1. Session management

2. Model and configuration

3. Context and memory

4. Tasks and code

5. Permissions and security

6. Remote and collaboration

7. Info and help

8. Other notes

• createMovedToPluginCommand.ts: a compatibility-shell factory for commands that moved to plugins (/pr-comments, /security-review use it)
• Feature-gated commands that do not exist in this snapshot: /proactive, /assistant, /fork (standalone edition), /workflows, /subscribe-pr, /torch, /peers, /buddy, etc.
• REMOTE_SAFE_COMMANDS (commands.ts:619): remote mode keeps only 17 purely local UI commands; BRIDGE_SAFE_COMMANDS (:651): phone/web can remotely trigger 6: /compact, /clear, /cost, /summary, /release-notes, /files
• In the external build, 19 .js command directories are dead-code empty shells (isEnabled:()=>false)

/clear — not just "clear the screen"

Entry clear/clear.ts:4 → conversation.ts:49 clearConversation(): 1. First run SessionEnd hooks (1.5s timeout, :68-74) 2. Report tengu_cache_eviction_hint telling the server to evict the prompt cache (:77-85) 3. Background agent tasks and main-session tasks (Ctrl+B) survive; foreground tasks are all killed (:93-107, 135-166) 4. setMessages([]) clears messages; update conversationId to force a redraw (:109-122) 5. clearSessionCaches() clears session caches (keeps surviving agents' private state), resets cwd, clears file-read cache / discovered skills / nested-memory paths (:124-132) 6. Reset AppState: file-history snapshots, attribution, MCP clients all reset (triggering reconnect); keep pluginReconnectKey (:168-191) 7. regenerateSessionId({setCurrentAsParent:true}) a new session ID (old ID recorded as parent for lineage analysis); surviving tasks' TaskOutput symlinks point at the new session directory (:203-224) 8. Re-persist mode/worktree state; run SessionStart hooks to inject messages (:230-250) In one sentence: a "soft restart" without restarting the process, accounting for lineage analysis, surviving background tasks, and the hook lifecycle.

/compact — dispatch a fork agent to write a summary

Entry compact/compact.ts:40: 1. Project away snipped messages; a user argument becomes a custom summary instruction (:46-52) 2. With no custom instruction, first try session-memory compaction (a cheaper incremental path :57-83) 3. Traditional path: first microcompact (in-place compact of old tool results), then compactConversation() (services/compact/compact.ts:387) 4. PreCompact hooks can inject extra summary instructions (mergeHookInstructions); streamCompactSummary (:451); on prompt-too-long, drop the oldest turns and retry, up to MAX_PTL_RETRIES (:460-491) 5. Fork-agent cache-sharing path (:1151-1203): runForkedAgent dispatches a sub-agent that runs only 1 turn (maxTurns:1) with no tools, carrying a system prompt / tools / message prefix identical to the main thread so it hits the prompt cache (comment: the old path had a 98% cache miss, wasting about 38B tokens a day); on failure, fall back to ordinary streaming (:1280-1326) 6. Summary template BASE_COMPACT_PROMPT (services/compact/prompt.ts:61): 9 fixed sections (Primary Request/Key Technical Concepts/Files and Code Sections/Errors and fixes/Problem Solving/All user messages/Pending Tasks/Current Work/Optional Next Step) 7. buildPostCompactMessages rebuilds the message sequence: compact boundary + summary + kept recent messages In one sentence: it does not delete messages; it "uses a cache-friendly fork sub-agent to write the old conversation into a structured summary, then rebuilds the session with that summary as the new foundation."

/init — an extra-long "interview script"

Entry commands/init.ts:226 (prompt type). The command itself does not scan — it returns a prompt and lets the main model do the work: - Old version (:6-26): have the model analyze the codebase and write CLAUDE.md, with a clear "should write" (build/lint/test commands, high-level architecture) and "should not write" (obvious advice, a per-file inventory) - New version (:28-224): an 8-stage script — P1 AskUserQuestion about building project/personal files; P2 dispatch a sub-agent to survey the codebase; P3 interview the user to fill blinds; P4/5 write files (every line passes a "if we deleted this, would Claude make a mistake?" test); P6 create skills; P7 create hooks; P8 summarize In one sentence: the extreme form of a prompt-type command — "the command is the script."

/memory — open an editor, not a conversation

Entry memory/memory.tsx:83: clearMemoryFileCaches() warms getMemoryFiles(), renders MemoryFileSelector; after a pick, mkdir -p, create with the wx flag, editFileInEditor() opens it with $VISUAL/$EDITOR (:21-58). Note: this snapshot has no # quick-memory prefix — inputModes only has !/ordinary etc.; the memory system = /memory command + memdir auto-memory + SessionMemory.

/model — picker + two-layer persistence

model/model.tsx: with no argument, render ModelPicker; setAppState updates mainLoopModel (in-memory :50-56) and updateSettingsForSource("userSettings",{effortLevel}) (ModelPicker.tsx:234) persists; with an argument, go through SetModelAndClose: isModelAllowed org allowlist → entitlement check → validateModel → persist; /model default restores the default.

/permissions — rule panel + "retry a denied operation"

permissions/permissions.tsx: PermissionRuleList five tabs (recent/allow/ask/deny/workspace); rules come from layered settings and are tagged with their source; applyPermissionUpdate takes effect in memory + persistPermissionUpdate writes the file; detects shadowed rules; onRetryDenials wraps recently denied commands as createPermissionRetryMessage and stuffs them back on the queue so the model retries.

/login — PKCE OAuth + local loopback listener

login/login.tsx:19 → ConsoleOAuthFlow (components/ConsoleOAuthFlow.tsx): 1. OAuthService generates a PKCE code_challenge+state, concatenates the authorize URL (services/oauth/client.ts:47-87) 2. Start a temporary localhost HTTP server listening on /callback (auth-code-listener.ts:10-16); after the browser authorizes, automatically capture the code and check state; if there is no jump in 3 seconds, show a manual-paste box (:157-181) 3. After swapping for a token, installOAuthTokens stores it in secure storage 4. Wrap-up (login.tsx:25-55): onChangeAPIKey() rebuilds the client, stripSignatureBlocks peels old thinking signatures, resetCostState(), refresh managed settings/feature flags, re-enrollTrustedDevice, authVersion+1 triggers a re-fetch of auth data In one sentence: standard OAuth2+PKCE, dual channel (loopback automatic / manual paste) so an SSH environment with no browser can still finish.

/resume — conversation archaeologist

resume/resume.tsx: with no argument, getWorktreePaths+loadSameRepoMessageLogs+filterResumableSessions render LogSelector (supports semantic search); after a pick, checkCrossProjectResume — same-repo worktree resumes directly; a different project is not resumed, instead a claude --resume ... command is concatenated and copied to the clipboard (:145-164); with an argument, first exact UUID match then search by title; what it reads is the ~/.claude/projects/<project>/<sessionId>.jsonl transcript file.

/cost — a purely local ledger

cost/cost.ts:6: subscribers see a subscription hint; API users formatTotalCost() (cost-tracker.ts:228) output total cost/duration/lines added-deleted/token usage grouped by model (including cache read/write); data all comes from in-process accumulators, no network request. Typical of a local command: a pure function formatting in-memory counters.

/help — render the command table as tabs

help/help.tsx returns <HelpV2 commands={commands}/>: two tabs, General (shortcuts) + Commands, iterating the full command set after getCommands() filtering (built-in + skills + plugins + MCP), formatDescriptionWithSource tagging the source. It has no static help text; it is a mirror of the runtime command registry.

/doctor — install-environment physical

doctor/doctor.tsx → getDoctorDiagnostic() (utils/doctorDiagnostic.ts:514): probe install type (npm-global/native/package-manager/development), version, multiple-install conflicts (gives a concrete uninstall command), auto-update permissions, ripgrep status, sandbox dependencies, shell aliases, context warnings. Entirely local, "a self-check script given a UI."

/review and /ultrareview — local and cloud twins

/review (prompt type, review.ts:34-47): LOCAL_REVIEW_PROMPT has the model itself run gh pr list/view/diff and output under "overview/code quality/improvement suggestions/risks" — inject an SOP. /ultrareview (local-jsx, review/ultrareviewCommand.tsx): checkOverageGate checks balance → launchRemoteReview creates a CCR remote session, sends the review prompt, registers a RemoteAgentTask that polls and pours cloud results back locally as a task-notification.

/commit — the textbook prompt command (internal-only in this snapshot)

commands/commit.ts:76: the prompt embeds four !`...` pre-executions: git status, git diff HEAD, git branch --show-current, git log --oneline -10 (:16-19) — executed locally before sending to the model, output substituted into the text; allowedTools whitelist only admits git add/status/commit; the prompt body is a "Git safety protocol" (do not change config, do not skip hooks, do not amend, do not commit suspected secrets) + a two-stage task (draft a message focused on why → HEREDOC a single message that stages+commits+signs off).

Load locations (skills/loadSkillsDir.ts:638-803)

Four sources: the managed-policy directory, ~/.claude/, the project directory (every .claude/ walking cwd up to home), --add-dir; plus plugin commands (loadPluginCommands.ts:414). Two formats: the new format .claude/skills/<name>/SKILL.md (directory only, :407); the old format .claude/commands/*.md (:566 tagged commands_DEPRECATED). Namespace: subdirectories become a colon prefix — .claude/commands/frontend/lint.md → /frontend:lint (buildNamespace :523-552); plugin commands carry a plugin:command prefix. Same names: first realpath wins (:725-763).

Template capabilities (getPromptForCommand :344-399)

1. Argument substitution (utils/argumentSubstitution.ts): $ARGUMENTS all arguments; $ARGUMENTS[0]/$0 the Nth; named arguments after frontmatter arguments: ticketId priority then $ticketId; with no placeholder, automatically append \n\nARGUMENTS: <args> (:140-142)
2. Shell pre-execution (utils/promptShellExecution.ts:69): two syntaxes, inline !`cmd` and block-level ```!; before send, executed by BashTool (or shell:powershell), stdout/stderr substituted into the text; execution first goes through hasPermissionsToUseTool; the skill's allowed-tools is temporarily merged into always-allow; safety red line: skills loaded from MCP always skip shell pre-execution (:371-374) — remote content is untrusted
3. Variable injection: ${CLAUDE_SKILL_DIR}, ${CLAUDE_SESSION_ID}
4. How it runs: default inline expansion into the current conversation; with frontmatter context:fork it runs as a sub-agent (executeForkedSlashCommand, processSlashCommand.tsx:62), optionally with agent: to specify the type

Full frontmatter field table (parseSkillFrontmatterFields :185-265)

description / allowed-tools / argument-hint / arguments / when_to_use / version / model (may inherit) / effort / disable-model-invocation (the model cannot invoke) / user-invocable (the user cannot type it) / hooks / context(fork) / agent / paths (conditional-skill glob) / shell / name

Full lifecycle of a prompt command

User types /my-cmd arg1 → getMessagesForPromptSlashCommand (processSlashCommand.tsx:827) → register skill hooks → addInvokedSkill records → generate a command-metadata message + isMeta expanded body → parse @file attachments → attach a command_permissions attachment → shouldQuery:true send to the model.