feat(agent): add dedicated crypto_agent for wallet & market ops (#1397)#1736
Merged
senamakel merged 3 commits intoMay 14, 2026
Conversation
…humansai#1397) Introduce a new built-in `crypto_agent` specialized for wallet actions (balances, transfers, swaps, contract calls) and market operations. The agent enforces a strict read → simulate → confirm → execute contract, with a narrow tool allowlist (wallet_*, stock_* market data, ask_user_clarification, memory_recall, current_time) — no shell, file_write, curl, or integration delegation. The safety preamble stays ON; identity / skills / memory boilerplate is omitted to keep the financial-risk voice tight. Routing: orchestrator gains `crypto_agent` in its `subagents` list so the runtime synthesises a `delegate_do_crypto` tool, and the orchestrator prompt explicitly routes crypto wallet/market requests here (never through `delegate_to_integrations_agent` or `delegate_run_code`). Coverage: - loader tests assert narrow allowlist, safety on, write-tool exclusions, iteration cap, and orchestrator routing entry. - prompt builder tests assert the read/simulate/confirm/execute contract, the prepared_id gating, and the no-fabrication / no-secret-logging rules are present. - capability catalog gains `automation.crypto_agent` for user-visible docs. Tool implementations themselves (wallet_*) land separately via tinyhumansai#1396 — unknown names are silently dropped by the tool filter today, so this PR is safe to merge ahead of them.
Contributor
📝 WalkthroughWalkthroughAdds a new built-in crypto agent: capability catalog entry, agent configuration and narrow tool allowlist, detailed prompt and builder enforcing read→simulate→confirm→execute, loader registration with tests, and orchestrator routing to a synthesized delegate for crypto operations. ChangesCrypto Agent Implementation
Sequence DiagramsequenceDiagram
participant User
participant Orchestrator
participant CryptoAgent
participant WalletTools
User->>Orchestrator: request involving wallet/market action
Orchestrator->>CryptoAgent: delegate_do_crypto (route request)
CryptoAgent->>WalletTools: read balances / inspect state
WalletTools-->>CryptoAgent: balances / positions
CryptoAgent->>WalletTools: wallet_prepare_transfer / quote (simulate)
WalletTools-->>CryptoAgent: quote + prepared_id
CryptoAgent->>User: confirmation prompt (fees/ETA/prepared_id)
User-->>CryptoAgent: user approval (ok to send)
CryptoAgent->>WalletTools: wallet_execute_prepared(prepared_id)
WalletTools-->>CryptoAgent: execution result or tool error
CryptoAgent->>User: final report (success or sanitized error)
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related issues
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/openhuman/agent/agents/crypto_agent/prompt.md`:
- Line 37: The prompt's "6. **On failure.**" rule currently instructs to "Show
the tool's error verbatim"; change it to require a sanitized/redacted error
instead: instruct the agent to present a short, faithful but redacted error
excerpt (remove/replace sensitive tokens, stack traces, request IDs, payloads),
then on the next line output the one-line likely cause (e.g. "RPC rejected —
nonce gap", "insufficient gas"), and then stop without retrying; ensure the
edited text explicitly forbids printing raw tool output and gives examples of
what to redact (secrets, full traces, IDs).
In `@src/openhuman/agent/agents/crypto_agent/prompt.rs`:
- Around line 17-45: The build function in prompt.rs lacks tracing around
entry/exit and section assembly; add debug/trace logs at the start and end of
PromptContext::build (function build) and after each section render call
(render_user_files, render_tools, render_safety, render_workspace) to record
presence/emptiness and length (e.g., trimmed length or "empty") and the final
output length; use the project's tracing/log crate at debug or trace level and
include contextual info like ctx identifier and ARCHETYPE usage so you can
correlate logs to specific prompt builds.
In `@src/openhuman/agent/agents/loader.rs`:
- Around line 522-579: The test currently checks a subset of the crypto_agent
tool allowlist/ban; extend the assertions in the match &def.tools block in
loader.rs to pin the full crypto tool contract by (1) adding assertions that
required grounding/utility tools like "stock_quote", "stock_exchange_rate",
"memory_recall", and "current_time" are present (similar to the existing
required checks for wallet_* and stock_crypto_series) and (2) adding assertions
that banned/dangerous tools such as "delegate_run_code" (and any other
disallowed entries mentioned in the contract) are NOT present (similar to the
existing forbidden loop). Reference the same match &def.tools block and reuse
the pattern of iter().any(...) assertions to tighten the test so TOML changes
cannot remove or reintroduce these tools silently.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 5530afd8-7597-41f2-a4d2-989aafc8662c
📒 Files selected for processing (9)
src/openhuman/about_app/catalog.rssrc/openhuman/agent/agents/crypto_agent/agent.tomlsrc/openhuman/agent/agents/crypto_agent/mod.rssrc/openhuman/agent/agents/crypto_agent/prompt.mdsrc/openhuman/agent/agents/crypto_agent/prompt.rssrc/openhuman/agent/agents/loader.rssrc/openhuman/agent/agents/mod.rssrc/openhuman/agent/agents/orchestrator/agent.tomlsrc/openhuman/agent/agents/orchestrator/prompt.md
) - prompt.md: sanitize tool error rule. The previous "show the tool's error verbatim" wording conflicted with the never-log-secrets rule earlier in the prompt — verbatim errors can carry payloads, request ids, stack traces, signed blobs. The on-failure step now requires a redacted summary plus the one-line likely cause. - prompt.rs: add debug/trace tracing around the build flow so prompt regressions are observable at runtime (entry with model/tool/skill counts, exit with prompt length + section-presence flags). - loader.rs test: pin the full crypto_agent tool contract. Previously the test only guarded `stock_crypto_series`; it now also asserts `stock_quote`, `stock_exchange_rate`, `memory_recall`, and `current_time` are present, and that none of `delegate_run_code`, `delegate_research`, or `delegate_plan` slipped into the allowlist via a future TOML edit.
Resolves a conflict in `orchestrator/prompt.md` where upstream's tinyhumansai#1731 added a new step-2 "live external service" branch and renumbered the decision tree (3 → 4). Keeps both upstream's integrations-first ordering and the crypto branch from this PR under the renumbered step 4 ("Does this need other specialised execution?").
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
crypto_agent(undersrc/openhuman/agent/agents/crypto_agent/) specialized for wallet actions and market operations.wallet_status/wallet_balances/wallet_supported_assets/wallet_chain_status,wallet_prepare_{transfer,swap,contract_call},wallet_execute_prepared,stock_quote/stock_exchange_rate/stock_crypto_series,memory_recall,ask_user_clarification,current_time. No shell, nofile_write, nocurl, no integration delegation.crypto_agentin itssubagents(synthesisesdelegate_do_crypto) and the prompt's decision tree routes crypto wallet/market requests there — explicitly not throughdelegate_to_integrations_agentordelegate_run_code.automation.crypto_agentfor user-visible docs.Problem
OpenHuman already supports built-in agent definitions under
src/openhuman/agent/agents/, but there was no domain-specific agent for crypto workflows. Routing wallet / trading work through generic agents either over-broadens the tool surface (shell, broad HTTP, etc.) or skips the read-before-write and quote-before-execute discipline a financial-risk agent needs. A dedicated agent gives the orchestrator a safe place to send these requests.Solution
src/openhuman/agent/agents/crypto_agent/withagent.toml,mod.rs,prompt.rs,prompt.md. Prompt builder mirrors the help/researcher pattern (archetype + tools + safety + workspace), keeps the global safety preamble ON, and omits identity / skills / memory boilerplate to keep the financial-risk voice tight.BUILTINSinagents/loader.rs; module declared inagents/mod.rs. Built-in count assertion bumped 15 → 16.src/openhuman/wallet/schemas.rsand the financial-apisstock_*series. Tools not yet registered are silently dropped by the tool filter at spawn time — so this PR is safe to merge ahead of Add crypto wallet tools for balances, transfers, swaps, and contract calls #1396.subagentslistscrypto_agentso the runtime auto-synthesises adelegate_do_cryptotool. The decision-tree section inorchestrator/prompt.mdadds an explicit branch for crypto requests.Submission Checklist
docs/TEST-COVERAGE-MATRIX.md(capability catalog updated instead).## RelatedCloses #NNNin the## RelatedsectionImpact
wallet_execute_prepared.Related
automation.crypto_agent,skills.wallet_execution,skills.connect_web3_wallet,skills.connect_crypto_exchangeAI Authored PR Metadata (required for Codex/Linear PRs)
Linear Issue
Commit & Branch
issue/1397-add-a-dedicated-crypto-transaction-agentc7c5e6029de2095bf1102ba587e85dc254bab0dbValidation Run
app/workspace changes,pnpm --filter openhuman-app format:checknot relevant.pnpm typechecknot relevant.cargo test --lib openhuman::agent::agents(54 passed),cargo test --lib openhuman::about_app(20 passed)cargo fmt+cargo checkclean for changed cratecargo fmt/cargo checkonapp/src-taurinot relevant.Validation Blocked
command:N/Aerror:N/Aimpact:N/ABehavior Changes
Parity Contract
stock_*tools).Duplicate / Superseded PR Handling
Summary by CodeRabbit
New Features
Documentation
Tests