dnd-alignment - SKILL.md Agent Skill

name: dnd-alignment description: Rank cobrabox features and pipelines on the D&D 9-alignment grid (Law/Chaos × Good/Evil). Use when the user invokes "/dnd-alignment", asks "what is the alignment of X", asks to "align my pipeline", or wants to know the moral character of a feature or analysis chain.

DnD Alignment

Determine the D&D 9-alignment of a cobrabox feature and append it to the canonical table.

Pipeline and roster rendering are handled by the Python script — do not compute them here.

Invocation

/dnd-alignment <FeatureName>

Responsibility split

Task	Owner
Rank an individual feature (assign Law/Good scores + lore)	This skill
Append ranking to the canonical table	This skill
Print the full roster	`uv run python -m cobrabox.egg.dnd_alignment --roster`
Compute pipeline aggregate	`uv run python -m cobrabox.egg.dnd_alignment F1 F2 F3`
Compute Chord pipeline aggregate	`uv run python -m cobrabox.egg.dnd_alignment --chord F1 F2 F3`

Canonical alignment table

The source of truth is src/cobrabox/egg/alignments.py.

Read that file before ranking a new feature — the feature may already be present.

Axis encoding

Law axis: Lawful = +1, Neutral = 0, Chaotic = −1
Good axis: Good = +1, Neutral = 0, Evil = −1

Label grid

             LAW (+1)          NEUTRAL (0)        CHAOS (-1)
GOOD  (+1)   Lawful Good       Neutral Good       Chaotic Good
NEUT  ( 0)   Lawful Neutral    True Neutral       Chaotic Neutral
EVIL  (-1)   Lawful Evil       Neutral Evil       Chaotic Evil

Procedure

1. Read the alignment table

Read src/cobrabox/egg/alignments.py. Check whether the feature is already present in ALIGNMENTS. If it is, print its existing entry and stop — do not re-rank.

2. Read the feature file

Read src/cobrabox/features/<feature_snake_case>.py to understand what the feature actually does before assigning an alignment.

3. Assign scores

Choose Law and Good scores using the rubrics below. Write 1–2 sentences of reasoning for each axis before committing to a score.

Law axis rubric

Score	Meaning	Indicators
+1 Lawful	Actively imposes structure	Creates segments (windowing), hard threshold classification (IQR), named category ontology (frequency bands), strict published-protocol adherence
0 Neutral	Passively describes existing patterns	Correlation/synchrony measures, spectral descriptions, statistical summaries — even with fixed formulas
-1 Chaotic	Disrupts or ignores conventions	`print` statements, missing validation, unpredictable output shape

Common trap: A fixed, deterministic formula does not make a feature Lawful — almost all signal processing is deterministic. Ask instead: does this feature impose structure onto the data, or describe structure already present in it?

Lawful examples: SlidingWindow (creates window segments), Bandpower (names frequency categories), SpikesCalc (classifies by IQR rule).

Neutral examples: Coherence, PLV, Autocorr, Spectrogram, EnvelopeCorrelation, PartialCorrelation — all use precise formulas but measure existing patterns without imposing.

Good axis rubric

Score	Meaning	Indicators
+1 Good	Preserves or enhances signal meaning	Increases interpretability, faithful to data, good metadata practice
0 Neutral	Indifferent to meaning	Mechanical reduction with no semantic intent (pure aggregation)
-1 Evil	Discards or distorts signal meaning	Selects extremes ruthlessly, drops metadata, lossy without documentation

4. Write one lore sentence

One punchy sentence (≤ 15 words) in the style of the existing entries. It should capture the moral character of what the feature does to data, not just describe it technically.

5. Choose a 2-char abbreviation

Use the first two lowercase letters of the class name, unless that conflicts with an existing abbreviation in the table — in that case pick the most recognisable 2-char substring.

6. Append to the table

Edit src/cobrabox/egg/alignments.py — add a new entry to the ALIGNMENTS dict following the existing format exactly:

"FeatureName": {
    "law":    <+1|0|-1>,
    "good":   <+1|0|-1>,
    "label":  "<Alignment Label>",
    "abbrev": "<2-char>",
    "lore":   "<lore sentence>",
},

Place it alphabetically by key, or at the end if alphabetical order is not obvious.

7. Run the roster to confirm

uv run python -m cobrabox.egg.dnd_alignment --roster

Confirm the new feature appears correctly in the output.

8. Report to conversation

Print:

<FeatureName>  —  <Alignment Label>
  Law:  <score>  (<reasoning>)
  Good: <score>  (<reasoning>)
  "<lore sentence>"

Table updated: src/cobrabox/egg/alignments.py
Run `uv run python -m cobrabox.egg.dnd_alignment --roster` to see the full grid.

Pipeline alignment — delegate to script

When the user asks about a pipeline or sequence of features, do not compute it yourself. Instead, tell them to run:

# Sequential pipeline
uv run python -m cobrabox.egg.dnd_alignment SlidingWindow LineLength MeanAggregate

# Chord pipeline (splitter + map + aggregator, framing features weighted ×2)
uv run python -m cobrabox.egg.dnd_alignment --chord SlidingWindow LineLength MeanAggregate

Roster — delegate to script

When the user asks for the full roster or grid:

uv run python -m cobrabox.egg.dnd_alignment --roster