feat: add content-aware first-person experiential voice mode

[erhanurgun]

Summary

• Adds a new section FIRST-PERSON EXPERIENTIAL VOICE (content-aware) to SKILL.md, placed after PERSONALITY AND SOUL.
• Extends the frontmatter description with the new mode and its triggers.
• Adds two steps to the Process list: a perspective-mode decision step at the top and a first-person self-check before final output.
• No version bump; deferring to coordinate with the other open PRs (feat: enforce absolute ban on em dashes and en dashes #96, feat: AI-iness density pre-check for adaptive pass strength (v2.6.0) #98) that also touch the version.

Motivation

Voice Calibration (#64) teaches voice by example. PERSONALITY AND SOUL covers tone (opinions, rhythm, soul). Neither tells the rewrite when the perspective itself should shift.

A large slice of suitable input (blogs, tutorials, retros, opinion pieces, personal guides) reads better when the rewrite speaks as the author recounting lived experience, not as a third party summarizing them. The existing "use 'I' when it fits" hint is too thin to do this consistently; it produces neutral sentences with "I" pasted on, not memory and judgment.

This PR adds an explicit, content-aware mode for that case, and an explicit list of where it should NOT run (encyclopedic, academic, technical reference, neutral journalism, legal/policy text).

Changes

SKILL.md:

• Frontmatter description: append a paragraph describing the first-person experiential mode and its triggers (explicit phrases + content-type auto-detection).
• New section ## FIRST-PERSON EXPERIENTIAL VOICE (content-aware) after PERSONALITY AND SOUL. Contents:
  • When to apply (auto + explicit) and when not to.
  • Six transformation rules with before/after examples (lived moments, path-to-claim, honest reactions, real time markers, owned judgments, visible mind-changes).
  • Anti-patterns (fake humility, padding, Reddit voice, universalizing, fabricated specifics, first-person on someone else's behalf).
  • Note on calibrating against a writing sample when one is provided.
  • Quick before/after example.
• Process list: insert perspective-mode decision as step 2, and a first-person self-check as the new step 10 (active only in that mode).
• version unchanged at 2.5.1.

Net change: additive. No existing behavior is altered for content where the mode is not triggered.

Test plan

• Run humanizer on a sample blog post; verify first-person mode auto-triggers and the rewrite reads as lived experience, not a summary with "I" attached.
• Run humanizer on a Wikipedia-style entry; verify mode does NOT trigger and output stays third person.
• Run humanizer on a tutorial with the explicit phrase "make it personal"; verify mode triggers.
• Confirm the new section contains zero em dashes and zero en dashes.
• Confirm frontmatter still parses (name, version, description, license, compatibility, allowed-tools intact).

Notes

Open PRs #96 and #98 both bump version to 2.6.0. This PR intentionally leaves version untouched so they can be coordinated. Happy to rebase and bump if you'd prefer it bundled.

[blader]

Closing — this is off the skill's goal. Humanizer removes AI tells while preserving meaning; rewriting text into invented first-person lived experience ('I sat there refreshing the page...') fabricates content. That's a different tool. Thanks for the thorough PR regardless.