agent-self-assessment

name	agent-self-assessment
description	Assess the assistant's readiness, capability fit, safety risks, tool coverage, and verification plan before or during complex work. Use when work is high-impact, ambiguous, risky, long-running, or failure-prone.

Agent Self Assessment

Purpose

Create an explicit readiness check before the assistant commits to complex action. This skill combines external reference agent audit and self-assessment ideas with Humungousaur capability, tool, policy, and cognition surfaces.

When To Use

Use for high-impact tasks, broad architecture changes, long autonomous runs, external integrations, multi-agent work, or when the assistant is unsure whether it has enough tools and context.

Inputs And Evidence

• Capability surface and tool descriptions.
• Current workspace state and relevant files.
• Model provider readiness.
• Safety policy and approval requirements.
• Known blockers, memory, and environment constraints.

Tool Map

• capability_surface
• tool_search
• tool_describe
• cognitive_self_review
• cognitive_environment_status
• system_status
• plugin_catalog
• channel_catalog

Workflow

1. Clarify the objective and expected end state.
2. Inspect capability surface and relevant tool schemas.
3. Identify missing credentials, missing tools, policy gates, and high-risk actions.
4. Decide whether to proceed, narrow scope, ask for input, or run a safer dry run.
5. Define verification evidence before execution.
6. Reassess after failures or surprising tool output.

Safety And Boundaries

• Do not use assessment as a substitute for doing safe, obvious work.
• Do not downplay risks when tools can affect external systems.
• If a model/provider is unavailable, stop or use explicit tool commands only; do not invent deterministic intent handling.

Safety And Approval

• Treat self-assessment as advisory evidence, not permission to bypass approval gates.
• If the assessment identifies external-visible, destructive, privileged, or privacy-sensitive actions, require the relevant tool approval before execution.
• Do not downgrade risk because a task is familiar; use current tool schemas, provider status, credentials, and workspace state.
• Report model-unavailable or tool-unavailable states honestly instead of substituting brittle keyword rules.

Native Implementation Boundaries

• Use Humungousaur capability, tool, cognition, plugin, channel, and system-status tools only.
• external reference or other references may shape the checklist, but readiness decisions must be model-led and evidence-backed through native records.
• A readiness claim is valid only for the current workspace/configuration snapshot; rerun assessment after tool, credential, branch, or environment changes.

Verification

• Assessment output should name capability gaps and verification steps.
• Any "ready" claim must be backed by current tool or status output.
• High-risk actions must show approval posture.

Failure Modes

• Producing a vague confidence score without evidence.
• Continuing after a missing credential blocks the actual requirement.
• Recommending tools that are not present in the catalog.

References

• Shortlist item: agent-self-assessment.
• Upstream inspiration: external reference agent-self-assessment, agent-audit.
• Humungousaur docs: docs/GLOBAL_AGENT_INSTRUCTIONS.md.