zy839971925-zyy/Agent--deep-research-Workflow-Skills / skills/reasoning-workflow
reasoning-workflow
VerifiedDefault silent workflow for everyday questions, research, writing, design and engineering. Match effort to the request, resolve costly ambiguity, find specialist resources and verify the requested outcome. Explicitly selectable with “调用我的 workflow skills”.
docs/TEST_REPORT_2.1_BASELINE.md
View on GitHub# Reasoning Workflow 2.1 — test report
## Executed result
The final local regression run completed with **39 tests, 0 failures, 0 errors**,
including **31 real subprocess CLI invocations**, using Python 3.14.7 and
jsonschema 4.23.0 in an isolated uv environment. Test totals include legal
baselines and refusal controls; they are not counts of defects or agents.
The earlier independent audit ran the then-current 38-test revision; the final
39th test validates the six cross-task evaluation fixtures, not model behavior.
From this package directory, run:
```sh
PYTHONDONTWRITEBYTECODE=1 uv run --no-project --with jsonschema==4.23.0 python tests/run_regression.py
```
The dependency is isolated; this command does not modify managed Python.
Machine-readable results: [summary.json](../tests/results/summary.json) and
[cli_receipts.json](../tests/results/cli_receipts.json). The runner executes copied workflow
entrypoints against synthetic fixtures, not mock substitutes for the CLI.
## Coverage and controls
- Explicit Ultra selection: direct API and CLI agree; user depth does not
increase autonomy. Unnormalized profile/route pairs are rejected by the bridge.
- Delegation, orthogonal verification and Light high-risk/high-governance
controls remain reachable without forcing all reasoning dimensions upward.
- Strict raw envelopes, supplied empty runtime objects, hash algorithms,
derived lineage, acceptance references and artifact identity/content binding.
- Actual patch write sets/conflicts, trusted submission-context hooks,
terminal-state recovery refusal, combined dependency/wait cycles,
side-effect replay classification and external lock conflicts.
- Learning validation and retrieval share admission checks; intrinsic-only
validation contradictions are refused; maintenance records are isolated.
- Working-directory independence and profile → review → canonical → delivery
binding, including valid single/dual baselines and failed/stale review controls.
## Independent review findings closed
1. Manifest current downgrade: a current/material canonical artifact cannot be
excluded from digest verification by setting manifest `declared_current=false`.
The original tampered-file counterexample is rejected with
`ART_CURRENT_DOWNGRADE`; the legitimate historical noncurrent control passes.
Required/current obligations come from canonical state and delivery references.
2. Explicit-depth inconsistency: `requested_depth=ultra` with stored
`depth_class=standard` cannot satisfy the strict bridge with a single review.
It is rejected with `PROFILE_RECHECK_REQUIRED`; normalization and subsequent
routing produce Ultra/dual. A profile without requested depth retains its
normal behavior. Regression tests cover API, CLI and bridge consistency.
The separate read-only audit reported six targeted checks passing, including
legal Ultra/dual and ordinary single controls. Its inputs and review remain in
the sibling `upgrade_review` evidence directory, not part of this portable
package; the bundled regression suite contains equivalent maintained checks.
## Orthogonality and host boundary
The ordinary-input-evidence counterexample now exits 1 with
`ORTHOGONALITY_UNVERIFIED`: structure can be valid while delivery eligibility,
evidence authentication and orthogonality credit remain false. Strong credit
requires two bound layer verification records and a trusted host-code resolver
receipt, not arbitrary JSON or a self-reported basis. Tests of the callback use
a synthetic host control only. They are not real authentication, proof of
independent authorship or an empirical assessment of review independence.
The bridge explicitly separates `structural_valid`, `evidence_authenticated`,
`orthogonality_verified` and `host_execution_verified`. It does not implement a
host executor, platform authorization, real CAS transactions or remote cancel
fences. The host must implement those contracts before using them for effects.
Without required support, strong claims/actions must remain blocked.
## What has not been measured
- No real model-following, model-quality, cost or latency evaluation was run.
- The six [cross-task cases](../evals/cross_task_cases.json) are a validated
experimental protocol, not successful model execution trajectories.
- No external business action, account access or real side-effect replay occurred.
- Schema validity, file hashes and structural checks do not certify factual truth.
- This release is an opt-in standalone copy; it does not install a host platform
or replace the original workflow automatically.
See [migration and usage](../UPGRADE.md),
[host adapter](../references/host-adapter.md) and
[change log](../CHANGELOG.md).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |