CtrlK
BlogDocsLog inGet started
Tessl Logo

uinaf/docs

Update repo documentation and agent-facing guidance such as AGENTS.md, README.md, docs/, specs, plans, and runbooks. Use when code, skill, or infrastructure changes risk doc drift or when documentation needs cleanup or restructuring. Do not use for code review, runtime verification, or boot/readiness infrastructure setup.

99

1.08x
Quality

100%

Does it follow best practices?

Impact

99%

1.08x

Average score across 3 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

structuring.mdreferences/

Structuring

Lightweight file shapes for agent work artifacts: specs, plans, and decisions. Templates, not ceremony.

Directory Layout

docs/
├── specs/           # WHAT + WHY — long-lived contracts
├── plans/           # HOW + WHEN — tactical, deleted when done
└── decisions/       # WHY we chose X — ADR-lite

These directories are not a replacement for reader-facing docs like README.md, CONTRIBUTING.md, architecture docs, or runbooks. Link them from routing docs when useful, but keep tactical planning out of the general documentation surface.

Specs

Contracts that define "done." Stable after agreement. Change only when requirements change.

Template

# <feature-name>

## Problem
One paragraph. What's wrong and for whom.

## Requirements
- R1: <requirement>
- R2: <requirement>

## Acceptance Criteria
- AC1: <observable, testable outcome>
- AC2: <observable, testable outcome>

## Constraints
- <what must be true: perf, compat, security>

## Non-goals
- <what this explicitly does NOT cover>

Naming

specs/<feature-slug>.md — e.g. specs/password-reset.md

Lifecycle

Write → review → agree → implement. Update only when requirements change, not when the plan changes.

Plans

Tactical execution strategy. Evolves during implementation. Delete when implemented — the branch/PR tells the story.

Plans are agent coordination artifacts, not public docs. Keep them narrow, link the owning spec when there is one, and remove them once the work lands unless the repo explicitly uses plans as an active backlog.

Template

# <feature-name> — Plan

Spec: `specs/<feature-slug>.md`

## Approach
2-3 sentences on the strategy.

## Tasks
- [ ] <task> → <files/modules>
- [ ] <task> → <files/modules>

## Order
What depends on what. What can run in parallel.

## Risks
- <what could go wrong and how to mitigate>

Naming

plans/YYYY-MM-DD-<feature-slug>.md — e.g. plans/2026-04-17-password-reset.md

Lifecycle

Write → implement → delete. The PR/commit history is the record. Plans are working documents, not archives.

Decisions

Why, not what. ADR-lite.

Template

# <NNNN>-<slug>

## Context
What forced the decision.

## Decision
What we chose.

## Consequences
What this enables and what it blocks.

Naming

decisions/NNNN-slug.md — e.g. decisions/0003-use-sqlite-not-json.md

Lifecycle

Append-only. New decisions get the next number. Add a new decision when you reverse course.

Discovery

Agents find these through AGENTS.md pointers, not filesystem scanning.

When the docs skill audits a repo and finds (or creates) specs/plans/decisions directories, it adds a routing entry to AGENTS.md:

## Docs
- Specs: docs/specs/
- Plans: docs/plans/
- Decisions: docs/decisions/

Keep AGENTS.md links shallow: directory plus one-line description, one pointer per directory.

Rules

  1. One purpose per directory. Keep specs in specs directories and plans in plans directories.
  2. Specs outlive plans. A spec survives multiple plan attempts. A plan is disposable.
  3. Delete plans when done. The PR tells the story. Plans are working documents, not archives.
  4. Link canonical docs. Plans reference specs by path. Decisions reference both.
  5. Drift is a signal. Plan changed but spec didn't = normal. Spec changed but tests didn't = bug.

SKILL.md

tile.json