xp-gate / delphi-review
AGENTS.md4 starsChanged 6 months agoMITView on GitHub
Generated: 2026-11-04 Commit: eb2097b0 Branch: sprint/2026-11-03-02 Version: 0.37.0.0
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository18 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/AGENTS.mdAGENTS.md
- githooks/AGENTS.mdAGENTS.md
- plugins/AGENTS.mdAGENTS.md
- skills/sprint-flow/AGENTS.mdAGENTS.md
- skills/test-specification-alignment/AGENTS.mdAGENTS.md
- admin-template-guidelinesSkill
- batch-grill-meSkill
- delphi-reviewSkill
- domain-modelingSkill
- grillingSkill
- grill-with-docsSkill
- improve-codebase-architectureSkill
- ralph-loopSkill
- sprint-flowSkill
- test-driven-developmentSkill
- test-specification-alignmentSkill
- to-issuesSkill
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 37
No hardcoded model lists — models defined by user's opencode.json configuration
- Distinctive: line 42
Do NOT terminate before achieving true consensus (≥90%)
- Distinctive: line 43
Do NOT reveal other experts' opinions during Round 1
- Distinctive: line 44
Do NOT accept partial agreement without resolution
- Distinctive: line 45
Do NOT auto-skip or bypass code-walkthrough based on change size.
- Distinctive: line 46
Do NOT degrade to single model on API errors (BLOCK)
- Very specific: line 47
Do NOT declare complete without writing .code-walkthrough-result.json
- Very specific: line 48
Do NOT treat provider: local fallback as an executed expert
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SKILLS/DELPHI-REVIEW KNOWLEDGE BASE
Generated: 2026-11-04 Commit: eb2097b0 Branch: sprint/2026-11-03-02 Version: 0.37.0.0
OVERVIEW
Delphi Consensus Review — multi-round anonymous expert review (≥90% threshold, exactly 3 experts with distinct executable model IDs). Supports design, requirements, and code-walkthrough modes.
STRUCTURE
skills/delphi-review/
├── SKILL.md # Core Delphi methodology + output contract
├── INSTALL.md # Setup instructions
├── AGENTS.md # This file
├── evals/ # Evaluation test cases
├── opencode.json.delphi.example # OpenCode delphi config example
├── references/
│ └── code-walkthrough.md # Code-walkthrough mode specification
└── .delphi-config.json.example # 3-expert config templateWHERE TO LOOK
| Task | Location | Notes |
| Core methodology | SKILL.md | Delphi process, expert roles, consensus rules |
| Code-walkthrough | references/code-walkthrough.md | Pre-push mode: complete review evidence with no hard file/LOC threshold |
| Config example | .delphi-config.json.example | 3 experts, distinct models |
CONVENTIONS
- 3 experts anonymous in Round 1 (no cross-expert bias)
- ≥90% consensus threshold (was 95%, now unified to 90%)
- Max 5 rounds before forcing decision
- Provider/vendor/nationality unrestricted; exactly three distinct trimmed requested model IDs are required
- Model selection: reads
delphi-reviewer-*agentmodelfields fromopencode.json - No hardcoded model lists — models defined by user's
opencode.jsonconfiguration - Code-walkthrough mode: triggered on git push, stores result in .code-walkthrough-result.json
- Code-walkthrough skipped on main/master pushes (by design)
ANTI-PATTERNS (THIS PROJECT)
- Do NOT terminate before achieving true consensus (≥90%)
- Do NOT reveal other experts' opinions during Round 1
- Do NOT accept partial agreement without resolution
- Do NOT auto-skip or bypass code-walkthrough based on change size; review large diffs completely or split them by user choice
- Do NOT degrade to single model on API errors (BLOCK)
- Do NOT declare complete without writing .code-walkthrough-result.json
- Do NOT treat
provider: localfallback as an executed expert
UNIQUE STYLES
- Anonymous expert reviews (Round 1)
- Statistical consensus measurement (≥90% threshold)
- Three modes: requirements review + design review + code-walkthrough
- Pre-push integration: .code-walkthrough-result.json stores commit hash + verdict
- Delphi guard in claude-code plugin: blocks Edit/Write before APPROVAL
COMMANDS
/delphi-review # Design review mode
/delphi-review --mode requirements # Phase 2 R1 requirements review
/delphi-review --mode code-walkthrough # Pre-push code walkthroughNOTES
- Used by pre-push hook for code-walkthrough validation
- Delphi guard (claude-code plugin) reads .sprint-state/delphi-reviewed.json
- Code-walkthrough result must match commit hash for verification
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.