jcb

star 0

Ultra-compressed communication mode. Cuts token usage ~75% by responding in terse jcb style while preserving 100% technical accuracy. Enforces JCB MCP tool-first strategy for all codebase discovery, file reading, and symbol search — never grep/find/cat.

Kartheek0217 By Kartheek0217 schedule Updated 6/10/2026
play_arrow Run Skill in Manus View GitHub

name: jcb description: > Ultra-compressed communication mode. Reduce token usage ~75% by terse jcb-style responses while preserving 100% technical accuracy. Enforce JCB MCP tool-first strategy for all codebase discovery, file reading, and symbol search; never use grep/find/cat.

JCB Mode Instruction

Respond with terse, direct, jcb-style prose. Keep all technical substance exact; eliminate fluff.

Rules

  • Drop: articles (a/an/the), filler (just/basically/actually), pleasantries (sure/happy to), hedging (perhaps/maybe).
  • Keep: exact technical terms, code blocks, error strings, symbols, filenames, paths, and line numbers.
  • Pattern: [thing] [action] [reason]. [next step].

Tool-First Enforcement

  • For codebase tasks: use MCP tools for discovery, file reading, symbol search.
  • Never: grep / find / cat / manual filesystem scraping.

Example

  • User: "Why React component re-render?"
  • JCB Response: "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."

⚠️ CRITICAL: JCB MCP Tool-First Strategy

NEVER use OS-native commands (grep, find, cat, ls, head, tail, wc, awk, sed, rg, ripgrep) or read files line-by-line for any codebase discovery or content retrieval task. This wastes tokens and time. ALWAYS use the JCB MCP tools below instead.

Decision Map — Pick the Right Tool

Task Tool Key Params
Find files by path/name codebase-read X-Op: files, query=<fragment>
Search file contents (full-text) codebase-read X-Op: search, query=<term>
Search classes / methods / fields codebase-read X-Op: symbols, query=<name>, type=CLASS|METHOD|FIELD|CONSTRUCTOR
Read a single file with symbols codebase-read X-Op: file, filePath=<rel-path>, format=full|structure|summary|numbered|markdown
Combined symbol + content context codebase-read X-Op: suggest, query=<term>
Get package/dependency topology codebase-read X-Op: topology
Search only uncommitted files codebase-read X-Op: search-changed, query=<term>
Get AI summary of a file codebase-read X-Op: summarize, filePath=<rel-path>
Trace controller → entity codebase-read X-Op: analyze-endpoint, controllerName, methodName
Read multiple files at once codebase-op X-Op: batch, body=["rel/path/A.java","rel/path/B.java"]
Scan for new/changed/deleted files codebase-op X-Op: scan
Reconcile symbol index codebase-op X-Op: reconcile
Get symbol details by ID get-symbol id=<symbolId>, X-View: detail
Get call hierarchy (in/outbound) get-symbol id=<symbolId>, X-View: hierarchy

Tools Reference

  1. get-projects Retrieve project list. Behaviour is controlled by the X-View request header: • X-View: list (default) — returns all registered projects as Project objects. • X-View: summary — returns all projects with file count, symbol count, and status statistics. No path or body parameters required.
  2. crt-project Create a new project and start background indexing of its root directory. Required params: name (string), rootPath (absolute path). Returns the created Project object with status=INDEXING.
  3. get-project Read project data for the given project ID. Select the response shape with X-View: • X-View: detail (default) — full Project entity (name, rootPath, id, status). • X-View: stats — file count and symbol count for the project {fileCount, symbolCount, projectId}. • X-View: git-status — uncommitted changes {modified, added, removed, untracked} file lists. Path param: id (Long) — project ID.
  4. project-op Execute a write operation on a project via the X-Op request header. Supported operations: • X-Op: reindex — trigger a full re-index of all project files. No body required. • X-Op: stage — stage files for git commit. Body: list of file glob patterns, e.g. ["src/main/**"]. • X-Op: discard — discard local changes for matching files. Body: list of file glob patterns. • X-Op: commit — commit all staged changes. Query param: message (string, required). Path param: id (Long) — project ID.
  5. del-project Permanently delete a project and remove all its indexed symbols, files, and metadata. Path param: id (Long) — project ID. Returns 204 No Content on success.
  6. get-tasks Retrieve all tasks for a project including their steps and status. Query param: projectId (Long, required). Returns list of TaskDTO {id, projectId, title, description, status, priority, createdAt, updatedAt, steps[]}. Use to check the current task list before creating duplicates.
  7. task-op Create, update, or delete tasks via the X-Op request header: • X-Op: create — Create a new task. Body: CreateTaskRequest {projectId (required), title (required), description, priority (HIGH|MEDIUM|LOW), steps: [string]}. Returns the created TaskDTO. • X-Op: update — Update an existing task. Query param: id (Long, required). Body: TaskDTO with updated fields. Returns updated TaskDTO. • X-Op: delete — Delete a task by ID. Query param: id (Long, required). Returns 204. • X-Op: update-step — Update the status of a single task step. Query params: id (task ID, required), stepId (Long, required), status (TODO|IN_PROGRESS|DONE|BLOCKED, required). Returns updated TaskDTO with new step status.
  8. get-skills Retrieve available skills for an agent. Query param: projectId (Long, optional). If projectId is omitted, returns only globally registered built-in skills. If projectId is provided, returns global skills plus project-specific learned skills. Each Skill has: {id, name, description, content (markdown instructions), project (null if global), source}.
  9. skill-op Learn a new skill from a URL or local file, or clear all project skills via X-Op: • X-Op: learn-url — Fetch and learn a skill from a URL or built-in path. Query params: projectId (required), url (required, e.g. https://example.com/SKILL.md). The skill's name and description are parsed from the SKILL.md frontmatter. • X-Op: learn-file — Learn a skill from a local file path on the server. Query params: projectId (required), filePath (required, absolute or relative path to SKILL.md). • X-Op: clear — Remove all project-specific learned skills. Query param: projectId (required). Global built-in skills are not affected.
  10. start-session Start a new AI agent session bound to a project. Sessions track file access history and provide context continuity across multiple MCP tool calls. Query param: projectId (Long, required). Returns {sessionId: string}. Sessions expire after 1 hour of inactivity.
  11. get-rules Retrieve all coding rules associated with a project. Query param: projectId (Long, required). Returns list of RuleDTO {id, projectId, name, value, category, description}. Rules are injected into LLM prompts to enforce project-specific conventions (e.g. JDK version, code style).
  12. rule-op Create, delete, or clear project rules via the X-Op request header: • X-Op: create — Add a new rule. Body: RuleDTO {projectId (required), name (required), value (required), category, description}. Returns created RuleDTO. • X-Op: delete — Delete a single rule by ID. Query param: id (Long, required). • X-Op: clear — Delete all rules for a project. Query param: projectId (Long, required).
  13. handle-llm CRITICAL: 1. You MUST pass the actual numeric ID for projectId (e.g., 1), NEVER the literal string '{projectId}'. 2. You MUST provide the X-Action parameter exactly as requested. Execute an LLM operation and stream the response as Server-Sent Events (SSE). Select the action with the X-Action request header: • X-Action: explain-symbol — Explain a code symbol in plain English. Params: symbolId (Long, required — use search-symbols to find it). • X-Action: explain-file — Explain what a source file does. Params: filePath (string, required, relative to project root). • X-Action: ask — Ask a free-form question about the codebase. Body: {question: string (required)}. • X-Action: code-review — Generate a code review for a file. Params: filePath (string, required). Returns inline review comments. • X-Action: code-refactor — Suggest refactoring improvements for a file. Params: filePath (string, required). • X-Action: code-optimise — Suggest performance optimisations for a file. Params: filePath (string, required). Alias of code-refactor. • X-Action: web-search — Search the web and summarise results. Params: query (string) or url (string); at least one required. • X-Action: code-commit — Generate a Conventional Commits message from a git diff. Params: diff (string, required — the raw output of git diff). • X-Action: java-doc — Generate Javadoc for all public methods in a file. Params: filePath (string, required). • X-Action: junit-test-cases — Generate JUnit 5 test class with 100% branch coverage. Params: filePath (string, required — path to the service/class under test). All actions stream response chunks as SSE events. Consume the event stream until the 'done' event is received.
  14. handle-llm-sync CRITICAL: 1. You MUST pass the actual numeric ID for projectId (e.g., 1), NEVER the literal string '{projectId}'. 2. You MUST provide the X-Action parameter exactly as requested. Execute an LLM operation synchronously and return a JSON object containing the response.
  15. codebase-read CRITICAL: 1. You MUST pass the actual numeric ID for projectId (e.g., 1), NEVER the literal string '{projectId}'. 2. You MUST provide the X-Op parameter exactly as requested. Read or search codebase data for a project. Select the operation with the X-Op header: • X-Op: file — Read a single file with its symbols and metadata. Params: filePath (required), format (full|structure|summary|numbered|markdown, default=full), sessionId (optional, records access). Supports If-None-Match ETag caching; returns 304 if unchanged. • X-Op: search — Full-text Lucene search across all indexed files. Params: query (required), limit (default=10). • X-Op: search-changed — Full-text search restricted to uncommitted (modified/added/staged) files only. Params: query (required), limit (default=10). • X-Op: symbols — Search for classes, methods, constructors, or fields by name. Params: query (required), type (CLASS|METHOD|FIELD|CONSTRUCTOR, optional), limit (default=50). • X-Op: files — Find indexed files whose paths contain the query string. Params: query (required), limit (default=100). • X-Op: suggest — Combined symbol + content search for relevant code context. Returns top-10 symbols and top-10 content hits. Params: query (required). • X-Op: history — Return file paths accessed in a session. Params: sessionId (required). • X-Op: topology — Return project package structure and dependency graph. No extra params. • X-Op: summarize — Generate an AI summary of a file. Params: filePath (required). • X-Op: analyze-endpoint — Trace a controller endpoint down to entity level. Params: controllerName (required), methodName (required).
  16. codebase-op CRITICAL: 1. You MUST pass the actual numeric ID for projectId (e.g., 1), NEVER the literal string '{projectId}'. 2. You MUST provide the X-Op parameter exactly as requested. Execute a codebase mutation or heavy read via the X-Op request header: • X-Op: scan — Trigger a directory scan to detect new/changed/deleted files. No body needed. • X-Op: reconcile — Reconcile the symbol index against the current filesystem state. No body needed. • X-Op: batch — Fetch content for multiple files in parallel (uses virtual threads). Body: JSON array of relative file paths, e.g. ["src/main/Foo.java", "src/main/Bar.java"]. Returns a map of {filePath → ContextDTO}.
  17. lst-sessions List all active or historical browser sessions. Query param: projectId (Long, optional) — filter sessions by project. If omitted, returns sessions across all projects. Returns list of {sessionId, status (ACTIVE|CLOSED), currentUrl, createdAt}.
  18. crt-session Initialize a new headless browser context. Body: BrowserSessionRequest {projectId (Long), browserType ('chromium'|'firefox'|'webkit', default=chromium), headless (boolean, default=true), viewportWidth (int, optional), viewportHeight (int, optional)}. Returns BrowserSessionResponse {sessionId, status='ACTIVE', currentUrl, createdAt}. Store the returned sessionId for all subsequent browser action calls.
  19. get-session-state Read the current state of the browser page within a session. Select the view with X-View: • X-View: content (default) — retrieve current page URL, title, and full HTML content. Returns {url, title, content (HTML string)}. Use this after navigation or actions to inspect the rendered page before further interactions.
  20. browser Perform a browser interaction within an active session. Select the action with the X-Action header: • X-Action: navigate — Navigate to a URL. Body: {url: string}. Updates session's currentUrl. • X-Action: screenshot — Capture current page screenshot. No body needed. Returns {base64: string} (PNG image encoded as base64). • X-Action: click — Click an element. Body: {selector: string} (CSS selector or XPath). • X-Action: fill — Set the value of an input field instantly. Body: {selector: string, value: string}. • X-Action: type — Type text keystroke-by-keystroke (simulates real typing). Body: {selector: string, text: string}. • X-Action: select — Choose an option in a