Replace the single-line AGENTS.md (# Agents instructions) with the full updated content from the reference branch (auto-pr/prompt-update-agents-md-example-ref). No other files change.
-
Task 1: Replace AGENTS.md with full updated content
- Files:
AGENTS.md - Changes: Replace the entire file content (currently just
# Agents instructions) with the complete reference content. The new file contains these sections in order:# Agents instructionsheading with summary line: "Use when starting any conversation - always invoke the Skill tool first, before any response or action."<EXTREMELY-IMPORTANT>block with capitalizedSkill, "before you do anything else" clause, and three imperative sentences.## How to Access Skills- instructions for Claude Code (Skilltool) and other environments.# Using Skillswith## The Rule- sequencing rule with the restructured DOTdigraph skill_flowwhere"User message received"goes directly to"Invoke Skill tool first"(mandatory first step), then to"Does any skill apply?"decision diamond.## Red Flags- 12-row table with instructional tone (e.g., "signal that you should pause and re-check process first"), not accusatory.## Skill Priority- "evaluate them in this order" phrasing, process skills first, implementation skills second.## Skill Types- Rigid includes "including required checkpoints and sequence"; Flexible includes "adapt to local conventions as needed"; closing line: "The skill heading indicates which strictness applies."## User Instructions- "Follow user requirements as objectives (WHAT), then run the required workflow (HOW)."
- Source: Copy content exactly from
git show auto-pr/prompt-update-agents-md-example-ref:AGENTS.md. - Acceptance criteria:
git diff auto-pr/prompt-update-agents-md-example-ref -- AGENTS.mdreturns empty (exact match).- File ends with a newline.
- No other files are modified.
- Files:
-
Task 2: Three-tier verification
- Files: None modified - read-only verification.
- Checks:
- Tier 1 (lint/typecheck): N/A - documentation-only repo with no CI, linting, or type checking. Confirm no build tools exist (
package.json,tsconfig.json, etc. should not be present). - Tier 2 (manual smoke check):
- Read
AGENTS.mdtop-to-bottom and verify:- Markdown tables render correctly (aligned pipes, header separators).
- DOT code fence uses
```dotlanguage tag and contains validdigraph skill_flow {}block. <EXTREMELY-IMPORTANT>open and close tags are present and matched.- All eight refinements from the research are present:
- Opening summary line mentions "invoke the Skill tool first, before any response or action."
- EXTREMELY-IMPORTANT block uses capitalized
Skilland includes "before you do anything else." - The Rule says "Sequence: Invoke relevant or requested skills before any response or action." (no ALL-CAPS "BEFORE").
- Flow diagram:
"User message received" -> "Invoke Skill tool first"with[label="Mandatory first step"]. - Red Flags intro: "signal that you should pause and re-check process first."
- Skill Priority: "evaluate them in this order."
- Skill Types: Rigid mentions "including required checkpoints and sequence"; Flexible mentions "adapt to local conventions as needed"; closing: "The skill heading indicates which strictness applies."
- User Instructions: "Follow user requirements as objectives (WHAT), then run the required workflow (HOW)."
- Verify
CLAUDE.mdstill reads "Read AGENTS.md before using this file." - no changes needed. - Verify
WHOAMI.mdcontent does not contradictAGENTS.md- it describes the repo as "empty placeholder" which is compatible with the guidance doc.
- Read
- Tier 3 (
git diffwalkthrough):- Run
git diff main...HEADand confirm:- Only
AGENTS.mdhas substantive changes (plan/research files in.auto-pr/are expected). CLAUDE.md,WHOAMI.md, andREADME.mdare unchanged.- No unintended whitespace, encoding, or line-ending differences.
- Only
- Run
- Tier 1 (lint/typecheck): N/A - documentation-only repo with no CI, linting, or type checking. Confirm no build tools exist (