========
Shred It
========

Run a deep adversarial critique of an idea, proposal, paper, grant, product
concept, or early research hypothesis.

.. note::

   Generated artifacts use the platform orchestration adapter. Claude can use
   agent teams where available; Codex and OpenCode ask before subagents and
   otherwise use the sequential persona fallback.

Description
-----------

The shred-it skill acts as a rigorous external reviewer. It maps the idea,
performs external research, checks prior art, tests feasibility and ambition,
looks for logic gaps, and produces concrete mitigations or kill criteria.

It is intentionally unsentimental. Praise is reserved for ideas that are
genuinely supported by evidence, novelty, feasibility, and impact.

Relationship to Other Skills
----------------------------

- **Pre-mortem** assumes a plan failed and explains why. Shred-it evaluates
  whether the underlying idea deserves to proceed at all.
- **Council** validates a plan from multiple perspectives. Shred-it adds
  external research, prior-art checks, novelty analysis, and iterative rebuttal
  handling.
- **Debate** argues two sides. Shred-it maps, researches, critiques, and
  proposes mitigations.

Triggers
--------

The skill activates when you mention:

- ``shred it``
- ``shred this idea``
- ``critique this idea``
- ``stress test this idea``
- external reviewer-style feedback

Input Format
------------

Shred-it accepts inline idea text or a markdown file with optional YAML
frontmatter.

.. code-block:: bash

   vi .claude/shred-it.md  # or .codex/shred-it.md / .opencode/shred-it.md

**Search order:**

1. Inline idea text in the user's request
2. Explicit path provided by the user
3. the platform default input file in the current repository root
4. legacy ``.claude/shred-it.md`` input
5. ask the user for an idea if no input is found

Frontmatter Fields
------------------

All fields are optional.

.. list-table::
   :header-rows: 1
   :widths: 20 10 20 50

   * - Field
     - Type
     - Default
     - Description
   * - ``reviewers``
     - list
     - *5 reviewers*
     - Reviewer definitions
   * - ``rounds``
     - int
     - ``2``
     - Number of inter-reviewer challenge rounds
   * - ``output_dir``
     - string
     - ``"resources/agent-docs/reviews/shred-it"``
     - Where reports are written
   * - ``context_budget_kb``
     - int
     - ``100``
     - Max total KB for embedded ``@file`` context
   * - ``research_depth``
     - string
     - ``"standard"``
     - ``quick`` / ``standard`` / ``deep``
   * - ``domains``
     - list
     - ``[]``
     - Optional topic domains
   * - ``previous_report``
     - string
     - ``""``
     - Prior shred-it report for iterative review
   * - ``response``
     - string
     - ``""``
     - User rebuttal or revised idea

Default Reviewers
-----------------

Shred-it uses five complementary reviewers:

.. list-table::
   :header-rows: 1
   :widths: 25 75

   * - Reviewer
     - Focus
   * - Prior-Art Researcher
     - Publications, patents, products, failed attempts, adjacent work
   * - Feasibility Reviewer
     - Resources, data, permissions, staffing, timelines, operations
   * - Logic and Evidence Critic
     - Causal leaps, undefined terms, unsupported claims, weak comparisons
   * - Ambition and Impact Reviewer
     - Novelty, contribution, scope, user need, incentive alignment
   * - Blockers and Mitigations Reviewer
     - Hard blockers, ethics, regulation, adoption, IP, kill criteria

Research Behaviour
------------------

External research is required unless the user explicitly requests an
offline-only critique. The skill prefers domain-specific MCPs or APIs where
available, then primary databases and official sources, then general web
search.

For biomedical, legal, financial, safety-critical, or policy ideas, the final
report should cite authoritative or primary sources and distinguish confirmed
facts from inference and speculation.

Workflow
--------

1. **Find and parse input**

   - Accept inline text or locate the markdown file
   - Parse frontmatter
   - Read prior report and response files for iterative reviews

2. **Embed local context**

   - Scan for ``@file`` references outside code fences
   - Read or condense files within ``context_budget_kb``

3. **Build an idea map**

   - Core claim
   - Mechanism or implementation path
   - Intended users
   - Required resources and permissions
   - Success metrics
   - Implied assumptions
   - Evidence-dependent claims

4. **Perform external research**

   - ``quick``: at least 5 targeted searches and 5 relevant sources
   - ``standard``: at least 10 targeted searches and 8 relevant sources
   - ``deep``: at least 15 targeted searches and 12 relevant sources

5. **Run reviewers**

   - Each reviewer critiques from a specialist perspective
   - Reviewers cite evidence, assign severity, propose mitigations, and set
     kill criteria

6. **Run challenge rounds**

   - Reviewers see each other's findings
   - They identify missed issues, weak evidence, duplicated concerns, and
     unjustified severity ratings

7. **Synthesize final report**

   - Verdict: ``PROCEED``, ``REVISE``, ``HOLD``, or ``KILL``
   - Research trail
   - Prior-art analysis
   - Critical issue register
   - Feasibility, ambition, logic, and blocker assessments
   - Mitigations and smallest decisive next tests
   - Iteration ledger when applicable

Iterative Reviews
-----------------

When ``previous_report`` and ``response`` are provided, shred-it preserves issue
IDs and marks each previous issue as:

- ``resolved``
- ``partially resolved``
- ``unchanged``
- ``worse``
- ``superseded``

A rebuttal only resolves an issue when it provides evidence or a concrete
design change.