Fragments

Fragments are composable, reusable quality-enhancing building blocks. Each fragment encapsulates a specific quality behavior — anti-fluff, objectivity, self-verification — and can be applied to any Context in a single call.

Fragments are the atoms that Blueprint orchestrates. They are designed to be layered: apply as many as you need, in any order.

Import

from mycontext.fragments import anti_fluff, objectivity, answer_first_fragment

Quick Start

from mycontext.templates.free.reasoning import RootCauseAnalyzer
from mycontext.fragments import anti_fluff, objectivity, self_check_analysis

# Build a context from a template
ctx = RootCauseAnalyzer().build_context(
    problem="API response times tripled after deployment"
)

# Layer on quality fragments
anti_fluff.apply(ctx)           # Concise, no filler
objectivity.apply(ctx)          # Objectivity rules in guidance
self_check_analysis.apply(ctx)  # Verification questions for analysis tasks

result = ctx.execute(provider="openai")

How Fragments Work

Each fragment carries:

• Constraints kwargs — fields like verbosity, answer_first, forbidden_phrases that are set on Context.constraints
• Guidance rules — behavioral rules appended to Context.guidance.rules

When you call fragment.apply(ctx):

1. Constraints fields are set only if not already present — existing values are never overwritten
2. Guidance rules are appended with deduplication — no duplicates

This means fragments are safe to layer: applying anti_fluff and then executive_brevity will not conflict, and user-set values always take priority.

Built-In Fragments

Fragment	What it does
anti_fluff	Sets verbosity="minimal", bans common AI filler phrases
answer_first_fragment	Sets answer_first=True — conclusion before reasoning
objectivity	Adds objectivity-enforcing rules to guidance
self_check_analysis	Adds analysis-focused verification questions (correlation vs causation, data gaps, confidence)
self_check_creative	Adds creative-focused verification questions (unconventional ideas, feasibility)
self_check_risk	Adds risk-focused verification questions (blind spots, worst case, mitigation)
structured_json	Sets output_contract for valid JSON output
grounding_strict	Adds strict evidence-grounding rules to guidance
educational_posture	Sets communication_posture="educational"
collaborative_posture	Sets communication_posture="collaborative"
hedging_ban	Bans hedging phrases ("it depends", "arguably", "it could be")
executive_brevity	Sets verbosity="minimal", answer_first=True, communication_posture="direct"

Using with Blueprint

Fragments are designed to work naturally with Blueprint-based architectures:

from mycontext.structure import Blueprint
from mycontext.foundation import Guidance
from mycontext.fragments import executive_brevity, grounding_strict

blueprint = Blueprint(
    name="executive_report",
    guidance=Guidance(
        role="Senior business analyst",
        rules=["Base every claim on provided data"],
    ),
    directive_template="Analyze: {topic}\n\nData: {data}",
    token_budget=3000,
    optimization="speed",
)

ctx = blueprint.build(topic="Q3 revenue decline", data=revenue_csv)

# Apply fragments for executive output
executive_brevity.apply(ctx)    # Minimal, answer-first, direct
grounding_strict.apply(ctx)     # Evidence-grounded claims

result = ctx.execute(provider="openai")

Creating Custom Fragments

from mycontext.fragments import Fragment

my_fragment = Fragment(
    name="medical_safety",
    description="Safety guardrails for medical-adjacent tasks",
    _constraints_kwargs={
        "forbidden_phrases": [
            "you should take",
            "I recommend this medication",
            "this will cure",
        ],
        "communication_posture": "educational",
    },
    _guidance_rules=[
        "Never provide specific medical diagnoses or treatment recommendations",
        "Always recommend consulting a licensed healthcare professional",
    ],
)

# Use like any built-in fragment
ctx = some_template.build_context(...)
my_fragment.apply(ctx)

Fragment API

Fragment

Field	Type	Description
name	str	Fragment identifier
description	str	What this fragment does

Method	Returns	Description
apply(ctx)	None	Merge fragment settings into Context in-place
constraints()	Constraints | None	Get the fragment's Constraints object
guidance_rules()	list[str]	Get the fragment's guidance rules list