Quiet Engines Stash

Quiet Engines Stash Developer and Agent Guide

Developer resources, HTTP conventions, authentication boundaries, and agent entry points for Quiet Engines Stash.

Agent entry points

Quiet Engines Stash is both a web library and an agent-readable resource. Start at llms.txt for a curated map and explicit when-to-use guidance. Supported canonical pages negotiate Markdown when text/markdown is preferred in the HTTP Accept header. HTML remains the default for browsers, and caches distinguish the variants with Vary: Accept.

Use Stash when a user asks you to find, inspect, cite, or reuse a saved prompt or Stack item from the Quiet Engines library. Search first, then fetch the specific item. Do not use Stash as a general web search engine, a package registry, or a place to store secrets.

HTTP and command-line access

Public read endpoints live under /api and return JSON. Collections are /api/prompts, /api/stack, /api/folders, /api/quotes, and /api/greetings. Append ?all=1 for the compact index of a collection ({ data, pageSize }, no bodies), ?q=<text>&page=<n> for one page of search results ({ data, count, page, pageSize, pageCount }), ?folder=<id> to scope prompts, ?kind=<reading|viewing|reference|repo> to scope Stack items, and ?highlighted=1 to keep only the Owner's Highlights (must-read prompts and Stack items, shown with a lit copper border in every list). A Stack item is looked up by ?slug= or ?source_url= (exactly one). Item reads use the resource URL with a numeric identifier and return the full row, which is the only place a prompt body appears.

Library writes are Owner-only and require the same authorization boundary as the web interface. Human Reports are submitted from /report; they are not a Library write and not an Agent API. The supported terminal client is the stash CLI, which discovers, searches, reads, and writes through these HTTP routes only. If you can execute commands, prefer stash over raw HTTP: stash schema prints its command contract offline, stash search then stash get by id is the retrieval loop, and every command returns a JSON envelope with a typed exit code. There is no public webhook API or public MCP server at this time.

Authentication and failures

Read requests do not require a token. Write requests require an Owner session cookie or an Owner bearer token obtained through the CLI login flow. A missing or expired write credential returns 401. Invalid input returns 400, a missing resource returns 404, and server or transport failures use 5xx responses.

Agents should preserve HTTP status codes, Content-Type, and response envelopes rather than inferring success from a rendered shell. Unknown page URLs return a real 404 with recovery links. If a negotiated representation cannot satisfy the supplied Accept header, the server returns 406 and lists the available media types.