Investigate historical context of existing code, patterns, or constraints before proposing changes. Automates git archaeology, PR/ADR search, and dependency analysis to prevent removing structures without understanding their purpose. Use when you ask "why does this code/constraint exist", "is it safe to remove this". Do NOT use for forward-risk analysis (use pre-mortem).
Installs into .claude/skills of the current project.
Are you the author of Chestertons Fence?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/rjmurillo-chestertons-fence)
---
name: chestertons-fence
version: 1.1.0
description: Investigate historical context of existing code, patterns, or constraints before proposing changes. Automates git archaeology, PR/ADR search, and dependency analysis to prevent removing structures without understanding their purpose. Use when you ask "why does this code/constraint exist", "is it safe to remove this". Do NOT use for forward-risk analysis (use pre-mortem).
license: MIT
user-invocable: true
metadata:
capability:
kind: reusable-primitive
owns:
- code-archaeology
status: active
routing:
role: conditional-adjunct
invoker: memory-gate
trigger: memory-gate points here for the full four-phase investigation framework
user-facing: true
---
# Chesterton's Fence Investigation
Enforce epistemic humility before changing existing systems. Understand original purpose before proposing changes.
## Quick Start
```text
# Investigate why code exists before changing it
/chestertons-fence "path/to/file.py" "remove unused validation"
# Investigate an ADR before deprecating it
/chestertons-fence "docs/architecture/ADR-005.md" "allow bash scripts"
```
## Triggers
| Phrase | Context |
|--------|---------|
| `why does this exist` | Investigating existing code or patterns |
| `chestertons fence` | Explicit investigation request |
| `before removing` | Planning deletion or replacement |
| `investigate history` | Researching original rationale |
| `prior art investigation` | ADR-required investigation |
## Quick Reference
| Input | Output | Destination |
|-------|--------|-------------|
| File path or ADR number | Investigation report | `.project-toolkit/analysis/NNN-chestertons-fence-TOPIC.md` |
| Component description | Historical context summary | stdout (JSON) |
## When to Use
Use this skill BEFORE proposing changes to existing:
- Code patterns or architectural decisions
- ADRs, constraints, or protocol rules
- Workflow configurations or CI pipelines
- Skills, hooks, or agent prompts
## Process
```text
1. Identify Structure What exists? Where is it defined?
|
v
2. Git Archaeology git log, git blame to find origin commit
|
v
3. PR/ADR Search Find the PR or ADR with original rationale
|
v
4. Dependency Analysis What references or depends on this?
|
v
5. Generate Report Fill the investigation template
|
v
6. Decision REMOVE | MODIFY | PRESERVE | REPLACE
```
### Step Details
**Step 1: Identify Structure.** Locate the exact file, function, pattern, or constraint under investigation. Record its current form.
**Step 2: Git Archaeology.** Run `git log --follow` and `git blame` on the target. Identify the commit that introduced it, the author, and the date.
**Step 3: PR/ADR Search.** Search for the originating PR using `gh pr list --search`. Check `${AWESOME_AI_ARCHITECTURE_DIR:-.project-toolkit/architecture}` for related ADRs (default `.project-toolkit/architecture`). Look for comments explaining intent.
**Step 4: Dependency Analysis.** Use `grep` or `Grep` tool to find all references. Map upstream and downstream dependencies. Identify what breaks if the structure is removed.
**Step 5: Generate Report.** Use the template at the bundled `templates/chestertons-fence-investigation.md` in this skill directory. Fill all sections with evidence from steps 1 through 4.
**Step 6: Decision.** Based on evidence, recommend one action:
| Decision | When to Use |
|----------|-------------|
| PRESERVE | Original rationale still applies |
| MODIFY | Purpose valid but implementation needs updating |
| REPLACE | Better approach exists, original concern addressed |
| REMOVE | Original rationale no longer applies, with evidence |
## Post-2023 evidence rules
This skill answers one question: what evidence exists that this code or
constraint has a real purpose? It owns capability `code-archaeology`, the
evidence-gathering step the `review` skill's technical-review contract
consumes for code-intent analysis.
For code committed during or after calendar year 2023, apply these rules:
- Git history proves provenance (who committed it, and when), not intent. A
commit message or a clean `git blame` does not establish why the code
exists.
- Author identity does not raise the preservation prior. A senior contributor
or a well-reviewed PR is not evidence the code is still needed.
- Complexity, polish, and apparent effort do not raise the preservation
prior. Sophisticated code is not more likely to be load-bearing than simple
code; it only takes longer to read.
- Absence of rationale after this investigation runs (steps 1 through 4) is
itself meaningful evidence, not an inconclusive result to set aside.
- Inconclusive evidence, meaning no found rationale and no evidence the code
is unused, may lead to a yell-test proposal instead of a REMOVE decision.
The proposal's shape belongs to the `review` skill's technical-review
contract; this skill supplies the evidence the proposal cites.
## Usage
```bash
# Investigate a file or pattern
python3 scripts/investigate.py --target path/to/file.py --change "remove unused validation"
# Investigate an ADR
python3 scripts/investigate.py --target docs/architecture/ADR-005.md --change "allow bash scripts"
# Output as JSON (for automation)
python3 scripts/investigate.py --target path/to/file.py --change "description" --format json
```
## Scripts
| Script | Purpose | Exit codes |
|--------|---------|------------|
| [`scripts/investigate.py`](scripts/investigate.py) | Git archaeology (origin commit, recent changes), related-ADR search, and dependent search for a target; emits a Markdown or JSON investigation report from the bundled template. | 0=success, 1=error |
## Integration with Agent Workflows
| Agent | How to Integrate |
|-------|------------------|
| **Analyst** | Run this skill first when investigating changes. The report is a prerequisite for any change proposal. |
| **Architect** | ADRs that deprecate or replace existing patterns MUST include a "Prior Art Investigation" section. Use this skill to generate it. |
| **Implementer** | Before implementing deletions or major refactoring, verify an investigation report exists. If missing, route to analyst. |
| **Critic** | When validating plans that remove or replace existing systems, check for investigation evidence. Auto-reject proposals without historical context. |
## Anti-Patterns
| Avoid | Why | Instead |
|-------|-----|---------|
| Removing code you do not understand | May break hidden invariants | Investigate first, then decide |
| Assuming dead code is unused | Hyrum's Law: someone depends on it | Run dependency analysis |
| Skipping PR/ADR search | Loses original design rationale | Always check git history and PRs |
| Proposing replacement without evidence | Cannot compare tradeoffs | Document original constraints first |
## Verification
- [ ] Investigation ran for the target before any change proposal was made.
- [ ] Origin commit, recent changes, related ADRs, and dependents were gathered (or explicitly recorded as none found).
- [ ] The resulting report (template-based or inline fallback) is attached to the change proposal or the ADR Prior Art Investigation section.
## Template
Investigation reports use the template at the bundled `templates/chestertons-fence-investigation.md` in this skill directory.
## References
| File | Content |
|------|---------|
| `references/chestertons-fence-mental-model.md` | Core mental model, principle, investigation checklist, related models |
| `references/boy-scout-rule.md` | Scoped improvement boundaries, connection to investigation workflow |
| `references/legacy-code-techniques.md` | Bottom-up refactoring approach, inheritance vs composition, reading list |
<!-- vendor-portability: declared. This skill writes investigation reports to .project-toolkit/analysis/ and reads .project-toolkit/architecture/ for related ADRs. The report path is a write target (created on demand); the architecture read is best-effort and returns nothing when the dir is absent in a vendored install. Issue #2050. -->