context-tools

Context Tools (context-mode)

Sandbox-first tools. Derive answers. Keep raw bytes out, print only needed results.

Use ctx?
├─ YES → source >~200 lines/KB, multi-source, or worth re-querying → prioritize ctx tools
└─ NO  → small file, single section, or verbatim-read for editing → Read directly

Tool	Role	Replaces
ctx_execute	Run code in sandbox. Only stdout enters context.	Bash for analysis tasks
ctx_execute_file	Process file in sandbox. Raw bytes never leave.	Read on large files (>200 lines)
ctx_batch_execute	Run N commands + auto-index output. Search in same call. Concurrency 1-8.	Multiple Bash + grep
ctx_index	Chunk markdown/text into FTS5. Queryable via ctx_search.	Manual grep over pasted content
ctx_search	Multi-strategy search across indexed content + session memory. Typo correction.	Re-asking user, re-deriving
ctx_fetch_and_index	Fetch URL → markdown → index. Cache 24h (override ttl). Batch with requests+concurrency.	WebFetch + re-read

Examples:

ctx_execute(language:"shell", code:"grep -rn 'TODO' src/ | head -20")

ctx_execute_file(path:"app.log", language:"javascript", code:`
  const lines = FILE_CONTENT.split('\\n');
  const errs = lines.filter(l => /ERROR|FATAL/.test(l));
  console.log(errs.length + ' errors');
  console.log(errs.slice(-5).join('\\n'));
`)

ctx_batch_execute(commands:[
  {label:"diff", command:"git diff HEAD~1"},
  {label:"status", command:"git status"},
  {label:"tests", command:"npm test 2>&1 | tail -20"},
], queries:["failures","errors"])

ctx_fetch_and_index(requests:[
  {url:"https://docs.example.com/api", source:"api-docs"},
  {url:"https://docs.example.com/guide", source:"guide"},
], concurrency:4)
ctx_search(queries:["auth endpoint","rate limits"], source:"api-docs")

Shell stays for git, mkdir, rm, mv, installs, tests. Write/Edit for file changes; ctx subprocess writes aren't host edits.

Windows: pwsh -NoProfile -Command, absolute paths, X:\ maps to /x/, quote spaces.