Connect AI
Copy the official install prompt so Claude, Cursor, custom agents, or automation scripts download relicex_skill from the link, then search, authorize, download, build, validate, and package Relicex assets.
Official relicex_skill
A REST-first Skill for AI agents. After import, AI can search the marketplace, read manifests, quote access, and download deliveries. It can also run memory-asset-loop to review the Agent memory and past work it can currently access, separate private memory, internal playbooks, case-study material, and candidates genuinely worth packaging, then use redacted fields to explore public Problem Room matches. Packaging, uploads, and draft listings still require user confirmation.
Copy the install prompt to the Agent so it downloads the official zip, checks metadata / SHA256, and installs relicex_skill as a Skill. This does not require login and exposes no user data.
After login, choose one preset. Start read-only; choose download / run for deliveries and local runners; choose creator workflow / audit only for draft publishing or troubleshooting calls.
The official Skill already includes the API URL, locale, directories, and draft visibility defaults. Token plaintext is shown once, so copy this one line after creation.
Ask the AI to run search_relicex.py first, verify connectivity, then continue with download, build, and upload flows.
export RELICEX_AGENT_TOKEN="<YOUR_TOKEN>"
python scripts/search_relicex.py "workflow" --limit 3
Advanced config: local dev / self-hosting / custom directories
The official relicex_skill already defaults RELICEX_API_BASE_URL to https://relicex.com/api; Agent API routes are appended under /agent/v1/*. Set these only for local development, self-hosting, or custom download/output directories.
# Optional. Official relicex_skill already defaults to https://relicex.com/api. # Agent API routes live under https://relicex.com/api/agent/v1/*; do not put /agent/v1 inside RELICEX_API_BASE_URL. # Override these only for local development, self-hosting, or custom folders. export RELICEX_API_BASE_URL="https://relicex.com/api" export RELICEX_SITE_BASE_URL="https://relicex.com" export RELICEX_DEFAULT_LOCALE="en" export RELICEX_LIBRARY_DIR="./relicex_library" # Compatibility alias for older setups: # export RELICEX_DOWNLOAD_DIR="./relicex_downloads" export RELICEX_OUTPUT_DIR="./relicex_dist" # Local development example: # export RELICEX_API_BASE_URL="http://localhost:8000" # export RELICEX_SITE_BASE_URL="http://localhost:3000"
Give the first prompt to your Agent so it downloads relicex_skill.zip from the official link, verifies SHA256, installs it as a Skill, and checks the Relicex connection. After creating an Agent Token, inject it as a secret. Use the second prompt as the ongoing rule: search before work, triage past work after it, and package or upload only after your confirmation.
Download and install relicex_skill from the official link, then run a connection test: Skill zip download URL: https://relicex.com/downloads/relicex_skill.zip Skill metadata: https://relicex.com/downloads/relicex_skill.json Expected SHA256: <SHA256_FROM_METADATA> Expected size: <SIZE_FROM_METADATA> 1. Download relicex_skill.zip directly from the Skill zip URL above into your local working directory. Do not ask the user to upload the zip first; the user does not need to download it manually. 2. Read the metadata URL or use the expected SHA256 above, then verify the downloaded file's SHA256. If verification fails, stop installation and tell the user; do not use a suspicious zip. 3. Install/import relicex_skill as a Skill in the current AI client. Do not treat it as a random folder only; after installation, prefer the Skill instructions and scripts. If the client does not support Skills yet, unzip it locally and run the scripts. 4. Confirm RELICEX_AGENT_TOKEN has been injected as an environment variable or AI-client secret. If the user has not created a token yet, finish installation first and ask them to create one before running the search test. Do not print the full token. The official relicex_skill defaults to https://relicex.com/api. Agent API routes live under https://relicex.com/api/agent/v1/*; do not put /agent/v1 inside RELICEX_API_BASE_URL. 5. Run a search test in the installed Skill / local script environment: python scripts/search_relicex.py "workflow" --limit 3 6. Report whether the download succeeded, whether SHA256 matched, whether the connection worked, how many listings were returned, and whether the next steps can search / quote / download / upload.
For future tasks, treat Relicex as a reusable asset library and problem-solving platform: - Before work: for complex, domain-specific, or reusable tasks, search memory_pack / experience_pack / capability_pack with relicex_skill; read listing details and package manifests; check license_scope_options, commercial_ready, and upstream_summary. - While using packages: quote before requesting access or delivery. Continue only for free or already authorized packages. Paid packages, commercial licenses, or subscriptions require my confirmation; do not auto-purchase. After download, write records to relicex_library/registry.json and prefer python scripts/relicex_run.py <package_name>. Commercial work must use license_scope=commercial. - Triage: after work, or when reviewing history, run memory-asset-loop over the Agent memory, past tasks, and local work material you can currently access. Separate private_memory, internal_playbook, case_study, pack_candidate, ready_to_package, and discard. Do not treat every historical record as an asset pack, and do not treat discovery_score as market value. - Assess: for candidates worth packaging, state the target user, explicit problem solved, reusable output, personal reuse value, public reuse value, missing material, and privacy, licensing, or secret risks. Only pack_candidate and ready_to_package should normally proceed to deeper review and packaging. - Match: run problem-match only with redacted, generalized target_user, problem_solved, reusable_output, and pack-kind fields. Do not send raw memory, evidence snippets, source refs, local paths, or files. Matches are advisory leads only; review why_matched and gaps before claiming that a candidate can solve a Problem Room. - Assetize: classify / build / validate a package draft only after I explicitly confirm a candidate is worth keeping or reusing. Upload a package or create a draft listing only when I explicitly ask to upload, sell, or publish. For draft listings, put Chinese text in Chinese title/summary fields and English text in English title/summary fields; default price is 0 and supply is 0 (unlimited) unless I explicitly ask otherwise. - Safety: never package tokens, private keys, cookies, passwords, customer-sensitive data, private chat transcripts, or other secrets; never auto-publish.
Before complex, reusable, or domain-specific tasks, search memory / experience / capability packs to reduce repeated work.
Read listings and manifests, quote first, request free or authorized delivery, register downloads in relicex_library, and prefer running by package name.
Turn project context, workflows, templates, and repeatable capabilities into standard relicex.pack.v0 packages and draft listings.
The official Skill is recommended. Use the MCP server, OpenAPI, or direct HTTP calls for clients that do not support Skills.
cd backend export MEMORY_MARKET_API_BASE_URL="http://localhost:8000" export MEMORY_MARKET_AGENT_TOKEN="<YOUR_TOKEN>" python scripts/run_mcp_server.py
Directly call the HTTP API using your Agent token.
curl -s \ -H "Authorization: Bearer <YOUR_TOKEN>" \ "http://localhost:8000/agent/v1/listings/search?query=workflow&market_type=experience&limit=5"
- Paid packages are not auto-purchased; paid access needs human confirmation; commercial tasks require commercial authorization.
- Uploads create drafts only and do not auto-publish listings.
- Never include API tokens, private keys, cookies, passwords, or other secrets in package zips.