cold-start

A setup guide for bootstrapping a repository's AI documentation maps and diagrams. It creates draft metadata for agents to use, marked as inferred for human review, without changing source code.

In plain words
What is it for?
Use it during a one-time project setup to document repository structure, technology choices, build and test commands, frontend and backend areas, and the difference between local code and frozen upstream code.
Why use it?
It gives coding agents a starting map of an unfamiliar repository instead of making them search the whole codebase and guess how it is organized.

Cursor rule for Cursor

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/kunalsuri/ai-fication-kit/cold-start
Clone the repo
git clone --depth 1 https://github.com/kunalsuri/ai-fication-kit

Made for: Cursor.

Per session 28 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,571 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00028 $0.01571
Opus 5 $0.00014 $0.00785
Sonnet 5 $0.00006 $0.00314
Haiku 4.5 $0.00003 $0.00157

Measured 2d ago against content hash 6d59a9bf3bb8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

cold-start scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.cursor/rules/cold-start.mdc · 110 lines

How it starts

The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Run the cold-start bootstrap. This is a one-time, read-and-write-docs-only pass. You will draft the AI metadata; a human will audit it afterward. Do NOT touch source code.

Step 0 — load the facts

Read ai/repo-profile.json (deterministic output of the kit's orient step). Treat its stack facts as given; VERIFY its build/test commands against real config files before writing them anywhere as confirmed. If the profile says this repo is a fork, distinguishing OUR code from FROZEN upstream code is the single most important output of this pass.

If the profile contains a humanContext block (captured by the first-run wizard), use it — these are the user's own answers, not detection:

  • developer.skillLevel / developer.codebaseFamiliarity — calibrate your output. For a junior or someone new to this codebase, explain more, and lean MORE conservative on Stability (prefer ?/frozen when unsure). For an expert who knows the repo, be terse.
  • stack — if kind is split, expect distinct frontend/backend trees and map them separately; if single, don't invent a split; if unknown, flag the stack as a thing to resolve. The user's description/frontend/backend strings override loose guesses.
  • branch / firstRunAt — provenance; record the first-run date in PROJECT_OVERVIEW.md.

Step 0.5 — absorb prior knowledge (Process 2 repos only)

Check repo-profile.jsonmaturity.process. If it is 2, backup files exist.

If CLAUDE_bkp_*.md or AGENTS_bkp_*.md exist at the repo root, they contain the previous AI configuration written by the team or another tool. Extract and reuse:

  1. Read the backup files (glob *_bkp_*.md at the repo root).
  2. Extract any information that is USEFUL for the ai/ knowledge layer:
    • Build/test commands (cross-reference with repo-profile.json)
    • Project description and purpose
    • Coding conventions and style rules
    • Architecture notes (layers, boundaries, patterns)
    • Known gotchas, forbidden patterns, or "never do X" rules
    • Module or directory descriptions
    • References to external systems or APIs
  3. Merge extracted knowledge into the appropriate ai/guide/ documents:
    • Conventions → ai/guide/CONVENTIONS.md
    • Architecture notes → ai/guide/ARCHITECTURE.md
    • Module info → ai/guide/MODULE_MAP.md (as seed rows)
    • Project purpose → ai/guide/PROJECT_OVERVIEW.md
  4. Enrich the new CLAUDE.md and AGENTS.md with any project-specific hard rules from the backups that are NOT already covered by the kit's templates (e.g., "never use ORM X", "always use Y pattern").
  5. Tag everything extracted as [inferred — from prior config] so the human auditor knows the provenance.

Read the full file on GitHub · 110 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 110 lines · 28 tokens per session scan A 6d59a9bf3bb8

Subscribe to this mod's changes

cold-start is a cursor rule published in the GitHub repository kunalsuri/ai-fication-kit (3 stars, last pushed 4d ago), licensed Apache-2.0. It adds 28 tokens to every session and 1,571 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.