http-fetch

http-fetch (sub-skill)

Direct shell wrapper for a single HTTP request. Replaces sub-agent sub-Agent steps that just GET/POST a URL — order-of- magnitude faster (no LLM round-trip, no tool surface, no iteration loop).

Inputs (with:)

key	required	default	notes
url	yes	—	absolute http(s) URL
method	no	GET	GET / POST / PUT / DELETE (case-insensitive)
body	no	''	request body, piped via stdin (for POST/PUT); empty = no body
timeout	no	30	request timeout in seconds
max_bytes	no	2000000	response body cap; larger payloads truncated + suffixed …

Output

• Success: response body on stdout (UTF-8 decoded, truncated to max_bytes if larger; lossy decode replaces invalid bytes).
• Non-2xx response: exit 1, stderr HTTP <code>: <reason> <body[:200]>; stdout still carries the body for callers that want to inspect.
• Network / DNS / timeout failure: exit 2, stderr cause.

When NOT to use

• Crawling multiple pages → use scrapling (via sub-agent).
• JS-rendered pages → use sub-agent + browser tools.
• OAuth dance / multi-step auth → use sub-agent.
• Streaming responses → not supported (we buffer + return).

Fallback

If this skill is unavailable, callers should spawn sub-agent with a curl/requests task — same result, ~10× the latency and a non-deterministic tool loop.