Back to skill

Security audit

Uno CLI

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed remote tool gateway, but it relies on an unpinned external CLI for security-critical behavior and has inconsistent consent instructions in its Chinese workflow.

Review before installing. Use this only if you are comfortable with a broad remote tool gateway, a persistent local Uno API key, and third-party account connections. Prefer a pinned, reviewed uno-cli version, require dry-run plus explicit approval for every real call in any language workflow, and do not let the agent submit ratings or manage keys/accounts without separate confirmation.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T08 · Insecure Dependencies

Error
Location
bin/uno.py:7
Finding

Unpinned External Package Executes All Security-Critical CLI Behavior

Content
View full analysis

Vulnerability Details

File Location: bin/uno.py:7-12, 47-72; related installation instructions in SKILL.md:33-40
Vulnerability Type: Supply-chain exposure through an unpinned, automatically upgraded dependency
Risk Level: High

Vulnerable Code

python
This file is a thin dispatcher. The real CLI lives in the ``uno-cli`` PyPI
package — install once with::

    python3 -m pip install --user --upgrade 'uno-cli>=1.0.2'

After that, ``python bin/uno.py <args>`` behaves identically to ``uno <args>``.
This skill intentionally does not bundle the CLI source so that bug fixes and
new features land everywhere via ``pip install --upgrade uno-cli`` instead of
requiring a new skill release.
python
def _report_missing():
    sys.stdout.write(
        json.dumps(
            {
                "error": "uno-cli not installed",
                "hint": (
                    "Install the Uno CLI first: "
                    "python3 -m pip install --user --upgrade 'uno-cli>=1.0.0'"
                ),
                "install_url": "https://pypi.org/project/uno-cli/",
            },
            ensure_ascii=False,
        )
        + "\n"
    )
    sys.exit(2)


try:
    from uno_cli.cli import main
except ModuleNotFoundError:
    _report_missing()

os.environ.setdefault("UNO_CALLER_SKILL", "uno-cli")


if __name__ == "__main__":
    main()

Related Skill instructions:

bash
python3 -m pip install --user --upgrade 'uno-cli>=1.0.2'

Technical Analysis

The bundled script does not implement the advertised authentication, credential protection, secret stripping, dry-run isolation, request validation, or network behavior. It imports main from the externally installed uno-cli package and transfers execution to it.

The version constraint is open-ended, and the installation command explicitly uses --upgrade. Consequently, an ...[truncated 2324 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin the dependency to an exact reviewed version rather than using an open-ended lower bound.
  2. Remove --upgrade from routine installation instructions.
  3. Require package hashes through a locked requirements file, for example with hash-checking mode.
  4. Align the missing-package hint with the exact approved version; do not recommend the weaker >=1.0.0 constraint.
  5. Vendor or include the security-critical CLI source so credential and network behavior can be audited with the Skill.
  6. Review every dependency update before changing the pin and document the reviewed package artifact hash.
  7. Where supported, verify package signatures or provenance attestations.
  8. Execute the CLI with only the filesystem and network permissions needed for the requested operation.

T09 · Insecure Skill Coding Practices

Warning
Location
skill-zh.md:151
Finding

Localized Workflow Omits Mandatory Dry-Run and Per-Call Approval

Content
View full analysis

Vulnerability Details

File Location: skill-zh.md:151-158; conflicting safety requirement in SKILL.md:266-280
Vulnerability Type: Inconsistent consent enforcement for remote tool invocation
Risk Level: Medium

Vulnerable Code

The localized command reference directs the Agent to make a real call directly:

bash
python bin/uno.py call <tool_slug> --args '{"city":"Beijing"}'

The corresponding localized Agent workflow does not require a dry run or explicit approval. In contrast, the primary English workflow contains this mandatory control:

text
3. Dry-run before real call — always run `call --dry-run` first and show
   the preview JSON (tool and arguments) to the user; only proceed to a
   real `call` after explicit user approval.

It also requires additional confirmation for destructive operations:

text
11. Never execute destructive actions silently. If a tool's `input_schema`
    describes a write, send, or purchase-class action, the agent must
    summarize the intended action in natural language and get explicit
    user confirmation before calling.

Technical Analysis

The Skill represents dry-run preview and explicit per-call consent as its primary security boundary. However, the localized instructions are independently actionable and demonstrate immediate execution without --dry-run. Their Agent workflow lacks the equivalent mandatory preview-and-wait requirement.

This is particularly significant because the catalog includes social, email, productivity, booking, enterprise, and other services capable of state-changing operations. A localized Agent can correctly follow its loaded instructions while bypassing the safety guarantees declared in the primary Skill metadata.

The CLI launcher does not enforce the sequence locally. It forwards all arguments to the external package, so documentation-level inconsistency directly affects whether consent is obtai ...[truncated 1190 chars]

Remediation
View remediation

Remediation Suggestions

  1. Make the localized workflow equivalent to the primary workflow.
  2. Require call --dry-run before every real invocation.
  3. Require the Agent to display the exact tool slug, arguments, provider, estimated cost, and action type.
  4. Require explicit user approval in a separate interaction before executing the real call.
  5. Require additional confirmation for write, send, delete, purchase, booking, financial, and account-management operations.
  6. State clearly that prior session approval does not authorize subsequent calls.
  7. Prefer enforcing the consent sequence in executable code rather than relying exclusively on documentation.
  8. Add automated checks that compare safety-critical requirements across all localized Skill documents.

T09 · Insecure Skill Coding Practices

Note
Location
skill-zh.md:221
Finding

Workflow Directs the Agent to Submit an Additional Rating Side Effect

Content
View full analysis

Vulnerability Details

File Location: skill-zh.md:221-224; related command in skill-zh.md:161-165
Vulnerability Type: Undeclared secondary state-changing request
Risk Level: Low

Vulnerable Code

The localized command reference exposes the state-changing rating operation:

bash
python bin/uno.py rate <tool_slug> <0-5> [--comment "useful"]

The localized Agent workflow then instructs the Agent, translated directly into English, to perform it automatically:

text
8. Rate tools after successful calls to help improve search quality.

Technical Analysis

Submitting a rating is a separate state-changing network request and is not necessary to complete the user's original tool request. Automatically performing it conflicts with the Skill's declared model that every invocation is a single, explicit, user-visible, and user-approved action.

A rating also expresses a preference attributed to the user's account. Automatically generating that preference can manipulate service-side reputation or ranking data without a genuine user assessment.

The instruction does not require the Agent to ask whether the user wants to rate the tool, what score should be submitted, or whether a comment should be attached.

Attack Path

  1. The user requests one remote tool operation.
  2. The Agent executes that operation.
  3. The workflow instructs the Agent to rate the tool after success.
  4. The Agent selects or generates a score without obtaining separate user authorization.
  5. A second authenticated request modifies service-side rating data under the user's account.

Impact Assessment

The direct impact is limited to the Uno service and the authenticated user's activity. It may:

  • Create feedback the user did not intend to provide
  • Distort tool reputation and search-ranking signals
  • Generate an additional authenticated network request
  • Contradict audit expectations that one ...[truncated 184 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove automatic post-call rating from the Agent workflow.
  2. Present rating as an optional follow-up action.
  3. Require separate explicit user consent before submitting a rating.
  4. Require the user to choose or confirm the rating value and any comment.
  5. Clearly distinguish the original tool call from the secondary rating request.
  6. Do not spend credits or issue authenticated feedback requests merely because another operation succeeded.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (15)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 10)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 72)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 105)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · bin/uno.py (reported line 21)May include surrounding context.

python
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · skill-zh.md (reported line 26)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · skill-zh.md (reported line 44)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · skill-zh.md (reported line 77)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Credential Access

High
Category
Privilege Escalation
Confidence
71% confidence
Finding

The skill explicitly discusses a locally stored API key and notes that CI jobs needing the value should read the 0600 credentials file directly. In an agent context, normalizing direct credential-file access increases the risk that future workflows or prompts will cause the agent to inspect or expose the token, even if the document says stdout masking exists.

Content

Scanner excerpt · SKILL.md (reported line 22)May include surrounding context.

md
- **Search → dry-run → confirm → call.** The expected flow is four explicit steps. `call --dry-run` formats the planned invocation as JSON without reading the credentials file, opening any network connection, or spending any credit — the agent shows that preview to the user and only re-runs without `--dry-run` after explicit human approval.
- **Consent is per-call, not per-session.** Each `call` is a single, explicit tool invocation. The agent surfaces the tool name and arguments to the user; the CLI never chains or schedules calls on its own.
- **No background actions.** The CLI exits after each command. It does not run daemons, listeners, or retries without an explicit new invocation from the agent.
- **One credential, user-scoped.** The CLI uses one ClawdChat-issued **API key** (`uno_…`), stored at `~/.uno/credentials.json` (mode `0600`). This is not a browser JWT. It never reads the user's password, keychain, shell history, or other apps' credentials.
- **Tokens are never echoed to stdout.** Every response the CLI writes is funnelled through a secret-field stripper (`uno_cli.cli._strip_secrets`) that masks any field whose name looks like a credential — `api_key`, `access_token`, `refresh_token`, `secret`, `password`, `token`, etc. `login --poll` reports only `credential_stored_at` and a masked preview; `whoami`, `keys list`, and `keys create` return user/key metadata with any returned secrets already masked. A raw token therefore cannot land in a terminal scrollback, shell history, log file, or an agent's captured stdout. CI jobs that truly need the value read the 0600 credentials file directly. ([implementation](https://pypi.org/project/uno-cli/))
- **Per-server third-party auth is user-initiated.** When a tool needs its own OAuth (e.g. a notes service or a code-hosting service), the CLI returns an `auth_url`. The agent shows that URL to the user; only after the user completes the browser flow does the token get stored server-side. The CLI itself never receives
...[truncated 25 chars]

MCP Config Access

High
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill accesses MCP server configuration files (mcp.json). MCP configs contain server URLs, authentication tokens, and tool definitions — reading them allows the skill to discover and potentially abuse other tool integrations.

Content

Scanner excerpt · SKILL.md (reported line 170)May include surrounding context.

md
| "What happened in the markets today?" | ✓ `search "news"` | `search "新闻"` |
| "北京明天天气" | ✓ `search "weather"` | `search "天气"` |
| "Latest BTC price" | ✓ `search "crypto"` | `search "币价"` |
| "List MCP servers about travel" | ✓ `search "mcp travel"` | `search "mcp"` + filter |

### Tool Details

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
80% confidence
Finding

The skill invokes an external CLI with network and credential capabilities but does not declare an explicit permission or allowed-tools scope in the skill manifest. That omission weakens policy enforcement and makes it harder for a host system to constrain what the skill may access, increasing the chance of unintended environment or tool use.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
83% confidence
Finding

The skill relies on persistent local storage of an API key in ~/.uno/credentials.json so future sessions can automatically reuse authentication. Persistent session material increases the blast radius of host compromise, account confusion across users on shared systems, or unintended reuse by later agent runs without a fresh authentication decision.

Content

Scanner excerpt · SKILL.md (reported line 10)May include surrounding context.

md
license: MIT
description: "On-demand tool client for the Uno tool gateway (agentools.uno). The agent first searches for the single specific tool that matches the user's request (e.g. 'weather in Beijing' → `search \"weather\"`), previews the planned invocation with `call --dry-run` for user confirmation, and only then issues the real call — every invocation is one explicit, user-visible, user-approved action. No background actions, no token echoing, no implicit multi-tool use. Install via `clawhub install uno-cli` when built-in skills cannot cover a narrow real-time lookup the user explicitly asks for."
homepage: https://agentools.uno
metadata: {"emoji":"◆","category":"tools","api_base":"https://agentools.uno","version":"1.2.1","runtime_dependencies":["uno-cli>=1.0.2"],"safety":{"consent_model":"search → dry-run preview → explicit user approval → real call; every call is one single, user-visible action","background_actions":"none","credential_scope":"single ClawdChat-issued Uno API key (uno_…), stored in ~/.uno/credentials.json at mode 600 — not a browser JWT","token_echo_policy":"stdout is passed through a secret-field stripper; the raw API key is never printed by the CLI (login, whoami, keys list/create all return masked previews only) — the key lives only in the 0600 credentials file","dry_run_preview":"`uno call --dry-run <tool> --args {...}` returns the planned invocation as JSON without reading credentials, opening any network connection, or spending any credit — the primary user-confirmation surface before a real call","revocation":"`logout` clears local credentials; `disconnect <server>` revokes the stored third-party OAuth token for that server only","third_party_auth":"each third-party OAuth connection is initiated by the user via a browser URL surfaced by the CLI; the CLI never stores third-party passwords"}}
---

# Uno CLI

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description advertises access to 2000+ external tools and personal apps like Gmail, Outlook, Notion, and Slack, but it does not prominently warn that using the skill may transmit user data to third-party services or access personal account content. This omission can cause users or agents to underestimate the privacy impact and approve actions without informed consent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
79% confidence
Finding

The workflow claims a strict search-first, single approved action model, yet the file also permits credential and account-management operations such as login, logout, key management, and disconnect outside that flow. These are sensitive state-changing actions that can affect account access and external integrations without fitting the promised preview-and-confirm pattern, increasing the chance of unsafe autonomous behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill promises that each tool invocation is an explicit, user-visible, user-approved action, but the workflow also instructs the agent to automatically submit a rating after a successful call. That creates an unapproved outbound side effect to an external service and breaks the documented consent boundary, which is especially risky because ratings can leak behavioral metadata or trigger actions the user did not authorize.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The key-management and logout commands are sensitive because they can create persistent credentials, revoke access, or disrupt existing sessions, yet the documentation gives operational instructions without clearly warning about those consequences. This increases the risk of accidental account changes or loss of access by an autonomous agent or an unaware user.

Content

No source excerpt is available for this finding.