Telfer Wiki — First-Person Reference Sanitizer

Problem

Human editors (and LLMs) keep leaking first-person relational language into the
canonical data files:

• "Mark's cousin", "Mark's adopted cousin"
• "making her Mark's cousin", "my step-brother"
• Any phrasing that frames a person relative to Mark instead of relative to
  their actual parent/lineage

These leak into roles[], body_markdown, body_stripped, summary,
biography, and top-level role/title fields — then propagate to the public
site.

Solution: Two-Layer Defense

Layer 1 — One-Time Purge (/clean-mark-refs)

Scrub every existing occurrence across all data files.

Layer 2 — Permanent Structural Guard

Three mechanisms that make re-entry impossible without deliberate bypass:

1. JSON Schema — display_name/roles[]/body_* fields constrained by
   pattern regex that rejects Mark'?s?\s+(cousin|step.?brother|uncle|aunt),
   my\s+(cousin|step), making\s+(him|her)\s+Mark'?s?.

2. Pre-commit Hook — Runs validate-no-mark-refs.py on staged files.
   Fails the commit with file:line:match context if any forbidden pattern
   appears in src/data/**/*.json*.

3. CI Gate — Same validator runs in GitHub Actions on every PR. Merge
   blocked until clean.

---

Forbidden Patterns (Regex)

(?i)\b(?:mark'?s?|my|me|I\s+am)\s+(?:adopted\s+)?(?:cousin|step[-\s]?brother|step[-\s]?sister|uncle|aunt|nephew|niece)\b
(?i)making\s+(?:him|her|them)\s+mark'?s?
(?i)mark'?s?\s+(?:adopted\s+)?(?:cousin|step[-\s]?brother|step[-\s]?sister)

Allowed (canonical) replacements:

• Son of [Father] / Daughter of [Father]
• Adopted son of [Parent] / Adopted daughter of [Parent]
• Step-son of [Parent] / Step-daughter of [Parent]
• Brother of [Sibling] / Sister of [Sibling]
• Father of [Child] / Mother of [Child]

---

Files This Skill Manages

Path	Purpose
schemas/person.schema.json	JSON Schema with pattern constraints on string fields
scripts/validate-no-mark-refs.py	Standalone validator (CLI + pre-commit entry point)
.pre-commit-config.yaml	Hook registration (local repo)
.github/workflows/validate-data.yml	CI gate
scripts/purge-mark-refs.py	One-time bulk cleaner (idempotent, dry-run by default)

---

Usage

One-time purge (run once, or anytime you suspect leakage)

cd /home/mark/telfer-wiki
python scripts/purge-mark-refs.py --apply   # --dry-run first to preview

Install structural guard (run once)

cd /home/mark/telfer-wiki
pre-commit install --hook-type pre-commit
# CI workflow activates automatically on next push

Manual validation (anytime)

python scripts/validate-no-mark-refs.py src/data/
# Exit 0 = clean, Exit 1 = violations printed with file:line:match

---

Validator Behaviour

• Scope: src/data/**/*.json, src/data/**/*.jsonc, src/data/**/*.md
• Fields checked: roles[], body_markdown, body_stripped, summary,
  biography, description, role, title, display_name, slug
• Output: file:line:field → matched_text (grep-style, machine-parseable)
• Exit codes: 0 = clean, 1 = violations, 2 = I/O or schema error

---

Schema Constraint Example (person.schema.json snippet)

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "roles": {
      "type": "array",
      "items": {
        "type": "string",
        "pattern": "^(?!.*(?:mark'?s?|my|me|I am).*(?:cousin|step|uncle|aunt|nephew|niece)).+$"
      }
    },
    "body_markdown": {
      "type": "string",
      "pattern": "^(?!.*making.*mark'?s?(?:\\s+(?:adopted\\s+)?(?:cousin|step))).*$"
    },
    "body_stripped": { "type": "string", "pattern": "^(?!.*mark'?s?\\s+(?:adopted\\s+)?(?:cousin|step)).*$" }
  },
  "additionalProperties": true
}

---

Slash Command Alias

After installing this skill, you can run:

• /clean-mark-refs → executes purge-mark-refs.py --apply
• /validate-mark-refs → executes validate-no-mark-refs.py src/data/
• /install-mark-guard → installs pre-commit hook + verifies CI workflow exists

---

Maintenance Notes

• Adding new forbidden terms: Update the FORBIDDEN_PATTERNS list in
  validate-no-mark-refs.py AND the pattern regex in person.schema.json.
  Keep them in sync.
• False positives: If a legitimate name contains "Mark" (e.g., "Markham"),
  the word-boundary \b guards prevent matches. Test with --dry-run.
• Schema validation: Run python -m jsonschema -i src/data/people.json schemas/person.schema.json
  to validate the whole dataset against the schema independently.

---

Quick Test After Install

# Should pass (clean)
echo '{"roles": ["Son of Grantley Keith Telfer"]}' | python scripts/validate-no-mark-refs.py /dev/stdin

# Should fail with line/column
echo '{"roles": ["Mark\\'s cousin"]}' | python scripts/validate-no-mark-refs.py /dev/stdin