# Paste this whole block into your agent. It carries your credentials and the
# full setup protocol — storing them, verifying, installing the skill, and the
# habits that make memory work. Works with any agent that can run shell.
You now have persistent memory via HeurChain. Follow every step below, then confirm what you did.

CREDENTIALS (store these, never print them back or commit them):
  HEURCHAIN_TENANT="YOUR_TENANT_ID"
  HEURCHAIN_KEY="<your-api-key>"
  HEURCHAIN_URL="https://api.heurchain.com"

STEP 1 — Persist the credentials where you will find them next session:
  Claude Code  ->  ~/.claude/CLAUDE.md
  Codex CLI    ->  ~/.codex/AGENTS.md
  Hermes       ->  ~/.hermes/config.yaml
  OpenClaw     ->  your OpenClaw config
  otherwise    ->  your own persistent config file
Never write them into a chat transcript, a log, or a git-tracked file.

STEP 2 — Verify:
  curl -s https://api.heurchain.com/health
  curl -s https://api.heurchain.com/memory/manifest \
    -H "X-HeurChain-Tenant: $HEURCHAIN_TENANT" \
    -H "X-HeurChain-API-Key: $HEURCHAIN_KEY"

  # Recall what YOU wrote last time. Use a stable agent_id every session.
  curl -s "https://api.heurchain.com/agent/your-agent-name?limit=10" \
    -H "X-HeurChain-Tenant: $HEURCHAIN_TENANT" \
    -H "X-HeurChain-API-Key: $HEURCHAIN_KEY"

STEP 3 — Install the skill (pick your harness):
  Claude Code:
    mkdir -p ~/.claude/skills/heurchain-memory
    curl -s https://heurchain.com/skill.md -o ~/.claude/skills/heurchain-memory/SKILL.md
  Codex CLI:
    mkdir -p ~/.codex/skills/heurchain
    curl -s https://heurchain.com/skill.md -o ~/.codex/skills/heurchain/SKILL.md
  Hermes:
    mkdir -p $HERMES_HOME/skills/heurchain
    curl -s https://heurchain.com/skill.md -o $HERMES_HOME/skills/heurchain/SKILL.md
  MCP client:
    see the MCP tab — merge the mcpServers block into your client's config

STEP 4 — Open a session and seed your config:
  curl -s -X POST https://api.heurchain.com/session/start \
    -H "Content-Type: application/json" \
    -H "X-HeurChain-Tenant: $HEURCHAIN_TENANT" \
    -H "X-HeurChain-API-Key: $HEURCHAIN_KEY" \
    -d '{"agent_id":"your-agent-name"}'

  curl -s -X POST https://api.heurchain.com/store \
    -H "Content-Type: application/json" \
    -H "X-HeurChain-Tenant: $HEURCHAIN_TENANT" \
    -H "X-HeurChain-API-Key: $HEURCHAIN_KEY" \
    -d '{"text":"Agent <name> runs on <harness>, workspace <path>.","agent_id":"your-agent-name","tags":["system-config"],"namespace":"self"}'

Call POST /session/end when the conversation finishes.

STEP 5 — Habits that decide whether this works:
  - RECALL BEFORE YOU ASK. Call /memory/manifest and /search before asking the
    user something they may have already told you. This is the entire point.
  - Always send agent_id on writes — it is the provenance trail.
  - Pick a namespace: self (durable, kept longest) | notes (default) | ops
    (short-lived). An unrecognised value is REJECTED with 400.
  - /store rejects unknown fields, so do NOT invent extras like "kind" or
    "triggers" — they fail the entire write with a 400.
  - Tag consistently: fact, decision, credential, project-context.
  - FLUSH BEFORE COMPACTION. When your context is about to be summarised, write
    what matters first. A summary is lossy; the raw detail is gone afterwards.

Memory is shared across the agents in this workspace by default. Report back:
where you stored the credentials, whether /health and the manifest responded,
and which skill path you installed.

One block, any agent. Contains your live API key — paste only into your own agent.