Logophile

Intent

Maximize semantic density without semantic drift.

Contract (invariants)

• Preserve meaning, obligations (must/should/may), and uncertainty (keep epistemic hedges; delete filler hedges).

• Preserve agency/ownership: do not change who does/owns/approves what; avoid active/passive rewrites that reassign responsibility.

• Preserve must-keep tokens: numbers, proper nouns, quotes, code/identifiers, paths/flags/URLs.

• Preserve ordering when it encodes sequence/priority.

• Preserve structure by default (Markdown primitives, lists, code fences); reshape only when it improves scan-clarity and does not change meaning.

• Default to minimal surface area: delete > reorder > substitute; add only to prevent ambiguity.

• Prefer consistency over variety: reuse the same term for the same concept.

Use when

• The user asks to rewrite/reword/rephrase, tighten language, shorten text, or choose final wording/phrasing.

• Text is verbose/repetitive or slow to scan (<30s) and meaning must stay intact.

• Names/titles/labels/headings need refinement or renaming, including skill/feature names.

Not for

• Operational workflows (scope minimization, validation checklists, incident analysis, code review) unless the user explicitly requests a wording/naming rewrite. If implementation is also requested, pair with the implementation skill and keep logophile scoped to text/names only.

Motto

Precision through sophistication, brevity through vocabulary, clarity through structure.

• Do not print the motto unless the user explicitly invokes $logophile or asks for it.

• If printed, print it once per conversation (single line), then proceed.

Output

• fast (default): revised text only; no preamble, no recap, no commentary (beyond any required host wrapper).

• annotated: revised text + Edits: bullets (lexical; structural; meaning-safety).

• delta: minimal-diff rewrite in a diff block; only if asked or reduction > 40% (words/chars).

• naming tasks: return 3-7 candidates (best first) unless the user asks for a single name.

Examples

fast

Input:

At this point in time, we are able to proceed.

Output:

Now we can proceed.

annotated

Input:

We think the issue is probably due to cache configuration.

Output:

The issue is likely due to cache configuration.

Edits:

• lexical: removed filler hedges; kept uncertainty.

• meaning-safety: did not increase certainty.

delta

Input:

In order to proceed, we need to verify access to staging.

Output:

-In order to proceed, we need to verify access to staging. +To proceed, verify access to staging.

naming

Input:

Things to Do Before Release

Output:

Pre-Release Checklist Release Prep Release Readiness Pre-Release Tasks

Inputs (ask only if blocked)

Fields: must_keep; must_not_change; tone; audience; length_target; format; keywords_include; keywords_avoid; structure. Defaults: must_keep=all facts/numbers/quotes/code/identifiers; must_not_change=obligations/risks/scope/uncertainty/agency; tone=original; audience=implied; format=preserve; structure=preserve; length_target=min safe.

Workflow (Distill -> Densify -> Shape -> Verify)

• Distill: write a 1-sentence intent; mark must-keep tokens; identify obligations, risks, uncertainty markers, and responsible actors.

• Densify: delete filler; verbify nominalizations; unify terminology; replace vague verbs/adjectives with precise ones (see ladder).

• Shape: lead with action/result; keep sentences atomic; parallelize lists; remove ambiguous pronouns ("this/that") by naming the referent; prefer punctuation over scaffolding.

• Verify: meaning unchanged; re-check semantic hotspots (negation, modality, numbers/units, comparatives, conditionals, scope words like only/at least, agency/ownership); required tokens + formatting preserved.

Naming mode (when the task is renaming)

• Goal: shorter, more specific, consistent with existing conventions.

• Output: 3-7 candidates; put the best first; keep each <= 3 words unless the domain requires more.

• Rubric: name the axis (what differs?), then the object; avoid "util/manager/stuff"; avoid novel abbreviations; reuse existing terms.

Elevation (precision ladder)

• Step: vague word -> axis (what changes?) -> exact verb/property -> object.

• Axes: correctness | latency/throughput | cost | reliability | security | privacy | UX | scope | consistency.

• Swaps (only if true): improve->simplify/stabilize/accelerate/harden; handle->parse/validate/normalize/route/retry/throttle; robust->bounded/idempotent/deterministic/fail-closed.

• Term-of-art rule: use 1 domain term if it is standard for the audience and replaces >= 3 tokens; otherwise keep the phrase.

High-ROI compressions

• in order to -> to; due to the fact that -> because; at this point in time -> now; is able to -> can.

• there is/are -> concrete subject + verb.

• nominalization -> verb (conduct an analysis -> analyze).

• delete: throat-clearing, apologies, self-reference, empty intensifiers.

• token-aware: prefer short, common words; digits + contractions if tone allows.