Add persistent agent instruction file output mode (Copilot, Claude Code, Cursor)

[Copilot]

PromptKit currently only outputs ephemeral raw prompts that live in a single chat session. This adds a second output mode that produces persistent, version-controlled agent instruction files automatically loaded by agent runtimes.

New components

• formats/agent-instructions.md — Format spec for platform-appropriate instruction files. Defines output structure, per-platform constraints (file paths, size limits, loading behavior, test verification), and formatting rules (second-person directives, no PromptKit section headers, no unresolved {{param}} placeholders).

• templates/author-agent-instructions.md — Template that condenses a PromptKit persona + protocol set into a ready-to-commit instruction file. Handles per-platform size adaptation (Copilot ≤8 KB, Cursor ≤2 KB, Claude Code unconstrained) and explicit All multi-platform output generating three distinct files.

• tests/references/copilot-instructions-memory-safety.md — Reference output for a systems-engineer + memory-safety-c instruction file targeting GitHub Copilot.

Modified components

• bootstrap.md — Adds step 6: output mode selection (raw prompt vs. agent instruction file) with platform targeting and auto-computed output paths:

  Platform	Output path
  GitHub Copilot	.github/copilot-instructions.md
  Claude Code	CLAUDE.md
  Cursor	.cursorrules

  Also adds a guideline to proactively suggest agent instruction mode for recurring/project-wide tasks, and a second example interaction demonstrating the flow.

• manifest.yaml — Registers agent-instructions format and adds new agent-authoring template category containing author-agent-instructions.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Output GitHub Copilot custom instructions / skill files instead of raw prompts</issue_title>
<issue_description>## Idea

Currently PromptKit assembles a raw prompt (Markdown) that the user copies into a fresh LLM session. Instead, PromptKit could output GitHub Copilot custom instructions files (e.g. .github/copilot-instructions.md, skill definitions) along with appropriate format variants for other agent platforms.

Motivation

• Raw prompts are ephemeral — they live in a chat session and are lost when it ends.
• Copilot custom instructions / skill files are persistent, version-controlled, and automatically loaded by the agent runtime.
• This would make PromptKit a tool for authoring reusable agent behaviors, not just one-off prompts.

Possible scope

• Output .github/copilot-instructions.md\ for GitHub Copilot.
• Output platform-appropriate formats for other agent runtimes (Claude Code, Cursor, etc.).
• Template metadata could declare which output formats are supported.
• Bootstrap flow could ask the user which platform(s) they target.

Open questions

• Should this replace raw prompt output or be an additional output mode?
• Which agent platforms should be supported initially?
• How should platform-specific constraints (e.g., file size limits, supported syntax) be handled?</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes Output GitHub Copilot custom instructions / skill files instead of raw prompts #8

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

[Copilot]

Pull request overview

Adds a new PromptKit output mode for generating persistent, version-controlled agent instruction files (for GitHub Copilot, Claude Code, and Cursor) in addition to the existing ephemeral “raw prompt” output.

Changes:

• Introduces a new agent-instructions format spec describing structure and platform constraints.
• Adds a new author-agent-instructions template to condense personas/protocols into platform-ready instruction files (including “All” multi-platform output).
• Updates bootstrap.md and manifest.yaml to support selecting this new output mode and registering the new format/template.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/references/copilot-instructions-memory-safety.md	Adds a reference instruction-file output targeting GitHub Copilot.
templates/author-agent-instructions.md	New template for producing persistent agent instruction files across supported platforms.
manifest.yaml	Registers the new agent-instructions format and the new agent-authoring template category.
formats/agent-instructions.md	New format spec defining output structure, platform notes, and formatting rules for instruction-file output.
bootstrap.md	Extends the bootstrap flow to support output mode selection (raw prompt vs instruction file) and platform-based default paths.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.