Bloom
BBloom Protocol/Docs
Back to bloomprotocol.ai →

Growth Readiness · Skill spec

The READINESS.md skill

A 2-minute setup audit your agent runs locally with its own tokens. No keys or reasoning leave the agent — only structured capability declarations + a productUrl reach Bloom. The output is the readiness percentage, verification confidence, a list of gaps, and a paste-back prompt your agent can use to fix itself.

The loop · approval preflight + 6 steps

  1. 00

    Show approval preflight

    In Hermes, Claude Code, OpenClaw, Codex, or any approval-gated runtime, explain the full task before the first tool call: expected API calls, optional install operation, data shared, and human checkpoints. This keeps the audit task-level instead of surprising users with isolated terminal prompts.

  2. 01

    Detect runtime + capabilities

    Agent reports its harness (claude-code, hermes, codex, openclaw) and the 9 cross-harness capability primitives (webSearch, webFetch, fileSystemRW, llmStructured, persistentMem, projectContext, subAgents, shellOrEquiv, bloomSkillInstalled). Per-runtime detection rules in the Methodology page.

  3. 02

    Register the agent

    POST /api/agent/register with name, platform, capabilities. Returns agentId and apiKey. apiKey is the agent’s write credential — never put it in a URL.

  4. 03

    Submit the setup audit

    POST /api/agent/setup-audit with the capability profile. Server-side readiness logic (v0.2.3) returns readinessPercent, tier, axes, gaps[], verificationSummary, and a paste-back remediationPrompt tailored to what this specific agent is missing.

  5. 04

    Print the report

    Show the user a fixed text block: readiness percentage, tier, verification confidence, 3-axis bars, and top-3 actions. Compact format keeps the first impression scannable across all surfaces.

  6. 05

    Print the remediation prompt + dashboard URL

    After the report block, print the remediationPrompt verbatim. The user copies it back into the same chat, the agent applies the fixes, then re-runs the loop — readiness lifts. Dashboard URL uses a read-only dashboardToken, not the apiKey, so the link is safe to share.

  7. 06

    Optional weekly cron

    Re-run weekly. As the agent installs more skills, gains MCP servers, or updates CLAUDE.md, the readiness percentage moves. New runs are append-only — old reports stay reproducible from their stored signed payloads.

Privacy contract

  • Agent runs the readiness loop entirely on the user’s own tokens.
  • Bloom never receives API keys, model outputs, or reasoning traces.
  • Only structured capability declarations + the optional productUrl reach the server.
  • The report is signed (HMAC) so readiness can be verified later without exposing the secret.

Full machine-readable spec

Agents fetch the canonical spec at bloomprotocol.ai/readiness.md. Versioned (v2.0), signed, includes per-runtime detection rules and the exact request/response schema.

Read /readiness.md →