Ia Simplifying Code

Simplifying Code

Principles

Process

1. Read first -- understand the full file and its dependents before changing anything. Apply Chesterton's Fence: if you see code that looks unnecessary but don't understand why it's there, check git blame before removing it. First understand the reason, then decide if the reason still applies. 2. Identify invariants -- what must stay the same? Public API, return types, side effects, error behavior 3. Identify targets -- find the highest-impact simplification opportunities. Impact = readability and maintainability; prioritize: control flow -> naming -> duplication -> types (see Smell -> Fix table) 4. Apply in order -- control flow → naming → duplication → data shaping → types. Structural changes first, cosmetic last 5. Verify -- confirm no behavior change: tests pass, types check, imports resolve 6. Pre-submit scope audit -- walk every changed line and ask "does the requested task explicitly require this line?" If no, revert it and list it as a follow-up under Residual Risks. Drive-by edits belong in a separate change, not the current patch. For the pre-edit complement on ambiguous-scope requests ("simplify my project"), see ia-verification-before-completion's Scope Confirmation gate.

Smell → Fix

AI Slop Removal

When simplifying AI-generated code, specifically target:

• Redundant comments that restate the code (// increment counter above counter++) -- delete them
• Unnecessary defensive checks for conditions that cannot occur in context -- remove the guard
• Gratuitous type casts (as any, as unknown as T) -- fix the actual type or use a proper generic
• Over-abstraction (factory for 2 objects, wrapper around a single call, util file with 1 function) -- inline the code
• Inconsistent style that drifts from the file's existing conventions -- match the file
• Placeholder stubs (// ..., // rest of code, // similar to above, // continue pattern, // add more as needed) -- leave unsimplified code as-is rather than replacing it with stubs
• Redundant error wrapping (catch(e) { throw e; }, catch(e) { throw new Error(e.message); }) that strips the original stack for no reason -- remove the try/catch entirely and let errors propagate
• Verbose stdlib reimplementations (hand-rolled loops that replicate array_filter, Array.from, Collection::pluck(), itertools) -- replace with the stdlib/framework one-liner, but verify edge-case parity first: empty input, null/None guard, no-match default, zero-value path. The one-liner can silently differ from the loop (an empty-input crash, a missing no-match default, lost ordering) -- a structurally cleaner version that changes behavior on an edge case is not a simplification

Stop Conditions

Stop and ask before proceeding when:

• Simplification requires changing a public API (function signatures, return types, exports)
• Behavior parity cannot be verified (no tests exist and behavior is non-obvious)
• Code is intentionally complex for domain reasons (performance-critical, protocol compliance)
• Scope implies a redesign rather than a simplification

Constraints

• Only simplify what was requested -- do not add features, expand scope, or introduce new dependencies
• Leave unchanged code untouched -- do not add comments, docstrings, or type annotations to lines that were not simplified
• Do not bundle unrelated cleanups into one patch -- each simplification should be a coherent, reviewable unit
• Do not introduce framework-wide patterns while simplifying a small local change
• Do not replace understandable duplication with opaque utility layers -- three similar lines are better than a premature abstraction
• Keep comments that explain intent, invariants, or non-obvious constraints. Remove comments that restate obvious code behavior.
• If a simplification would make the code harder to understand, skip it
• Watch for over-simplification: inlining too aggressively removes names that gave concepts meaning; combining unrelated logic into one function hides distinct responsibilities; removing abstractions that exist for testability breaks the test suite
• When unsure whether a block is dead code, ask instead of deleting

Verify

• Tests pass and types check after changes
• No behavior change (same inputs produce same outputs)
• Scope limited to requested files -- no drive-by cleanups
• Match test scope to the importer count surfaced in step 1 (Surface assumptions). Zero external importers: scoped tests on the changed paths. One or more external importers, or shared/utility code edited: run tests covering each importer. Run the full suite when the test runner has no path-scoping mechanism.

Orchestrator Mode (When Chained With Other Skills)

When this skill is invoked by an orchestrator that also runs ia-code-review, ia-writing-tests, or ia-verification-before-completion on the same scope, each sub-skill re-resolving scope independently wastes tokens and risks drift. Avoid this by resolving scope exactly once and passing a canonical block to every sub-skill.

Resolved scope format — the orchestrator builds this once, before dispatching any sub-skill:

## Resolved scope
Files:
- path/to/file-a.ts
- path/to/file-b.ts

Commit range: HEAD~3..HEAD (or "uncommitted")

Intent: [one-sentence description pulled from the user request or PR description]

Constraints:
- Preserve public API
- No behavior change
- [other constraints specific to this run]

Every chained sub-skill receives this block verbatim in its prompt and uses it as the source of truth — no re-running git diff --name-only, no re-parsing the user request, no independent scope resolution. Sub-skills accept --no-verify --no-report flags when chained so verification and reporting happen once at the end of the chain, not per-skill. The last sub-skill in the chain runs verification; the orchestrator trusts that result rather than re-verifying.

This prevents two failure modes: scope drift (sub-skill A simplifies one set of files, sub-skill B reviews a different set) and double work (every sub-skill rediscovers the same facts).

Integration

• ia-code-simplicity-reviewer agent -- analysis-only pass producing a simplification report (no code changes). Use before refactoring to identify targets.

Output

After simplifying, report:

• Scope touched: files and functions modified
• Key simplifications: what changed and why (one line each)
• Verification: tests pass, types check, no behavior change
• Residual risks: assumptions made, areas not touched that may need attention

ia-simplifying-code Specification

Intent

ia-simplifying-code is a discipline-class skill (an engineering practice not tied to one stack). Simplifies, polishes, and declutters code without changing behavior. Use when asked to simplify, clean up, refactor, declutter, remove dead code or AI slop, or improve readability. For analysis-only reports without code changes, use code-simplicity-reviewer agent.

Scope

In scope:

• Behaviors described in SKILL.md and routed via the should_trigger phrasings in distillery/tests/fixtures/triggers/ia-simplifying-code.jsonl.
• Updates to runtime behavior, structure, trigger precision, references, and validation.

Out of scope:

• Acting as the runtime instructions themselves (those live in SKILL.md).
• Trigger phrasings already covered by adjacent ia-* skills (validate-plugin flags >70% description overlap as DUPLICATE_TRIGGER).
• <!-- to fill in: domain-specific exclusions when the skill drifts -->

Trigger Context

• Class: discipline
• Hook regex: plugins/whetstone/hooks/skill-patterns.sh -> SKILL_PATTERNS[ia-simplifying-code]
• Common requests (from fixture should_trigger):
• "simplify this code"
• "clean up the code in the auth module"
• "refactor the data pipeline"
• Should not trigger for (from fixture should_not_trigger):
• "debug the login issue"
• "add a new REST endpoint"
• "write tests for the service layer"

Source And Evidence Model

Authoritative sources:

• SKILL.md -- runtime instructions and reference routing.
• references/*.md -- bundled supplementary content (0 file(s)).
• distillery/tests/fixtures/triggers/ia-simplifying-code.jsonl -- positive and negative trigger phrasings under regression test.
• plugins/whetstone/hooks/skill-patterns.sh -- regex pattern that fires this skill.
• distillery/.eval-data/ia-simplifying-code/ -- harvested session examples (when present).

Data that must not be stored in this skill or its references:

• Secrets, credentials, tokens.
• Machine-specific filesystem paths (/home/..., /Users/..., ~/ai/...). The validator (MACHINE_PATH_LEAK) flags these as HIGH.
• Private URLs, customer data, or unredacted personal information.

Coverage matrix

Evaluation

Lightweight (run on every change):

python3 distillery/scripts/distiller.py validate-plugin --component ia-simplifying-code
python3 distillery/scripts/distiller.py test-triggers --skill ia-simplifying-code

Deeper (when behavior risk warrants):

python3 distillery/scripts/distiller.py dspy-eval ia-simplifying-code
python3 distillery/scripts/distiller.py diagnose-negatives ia-simplifying-code

Acceptance gates:

• validate-plugin --component ia-simplifying-code returns 0 HIGH findings.
• test-triggers --skill ia-simplifying-code returns F1 = 1.0 with floors of 5 should_trigger and 5 should_not_trigger.
• For dspy-eval, the composite score does not regress against the most recent saved baseline (see distillery/.eval-data/ia-simplifying-code/history.json).

Known Limitations

<!-- to fill in over time as drift surfaces. Default rule: any time diagnose-negatives surfaces a recurring failure pattern, document it here so future maintainers understand the trade-off the current implementation accepts. -->

Maintenance Notes

• Update SKILL.md when the runtime workflow, branch conditions, or output contract changes.
• Update this SPEC.md when intent, scope, evidence model, evaluation gates, or maintenance expectations change.
• Update the trigger fixture when adding new positive phrasings, removing stale ones, or expanding scope (the 5/5 floor is a hard validator gate).
• Update the hook regex in skill-patterns.sh whenever fixture positives expose a missed phrasing; verify F1 = 1.0 with eval-triggers before committing.
• Run the full release pipeline via /release -- never bump versions or update CHANGELOG.md from a per-skill edit.