with-agents

Your Repo Is the Memory: Durable Context for AI Coding Agents (Image Deck)

An image-only alternate deck for comparing generated presentation slides against the Markdown-native durable-context deck.

AI AgentsWorkflowsAgent Skillsdot-agents
→ View as slides

Your Repo Is the Memory

Generated cover slide for Your Repo Is the Memory, with the subtitle Chat is RAM and the repo is disk.

Opening slide for the generated image-only deck.

Coding agents made implementation cheap

Generated slide saying coding agents made implementation cheap and asking where the memory should live.

Use the new generated slide sequence in file order.

Start with the checkpoint you already trust

Generated slide showing a checkpoint timeline from prompt to incident, with the trusted checkpoint moving earlier.

Fast form without shared intent

Generated slide saying generative AI makes form cheap, with documents and code artifacts spreading without shared intent.

The PR inherits too many jobs

Generated slide comparing the old plan-discuss-build-review path with the agent path where alignment moves onto the critical path.

Speed becomes comprehension debt

Generated slide showing fast generation, large diffs, shallow review, and green tests turning speed into comprehension debt.

The model is not your project memory

Generated slide contrasting broad lossy model memory with exact local repo truth.

The repo is disk; chat is RAM

Generated slide showing messy chat distilled into durable repo-local context that preserves the parts that change future behavior.

What should survive?

Generated slide showing messy chat exploration filtered into repo-local context that changes the next run.

The repo-memory stack

Generated slide mapping durable repo memory into AGENTS.md, skills, work items, decisions, and evidence.

AGENTS.md is the map

Generated slide explaining AGENTS.md as the map for stack, directories, ownership boundaries, checks, and rules.

Agent Skills are procedures

Generated slide defining a skill as trigger, steps, scripts or assets, and proof target.

dot-agents keeps task state alive

Generated slide showing a dot-agents work item with index, research, plan, progress, and decisions files.

Keep task-local context close to the task

Generated slide showing a work item surviving across sessions, preserving progress, decisions, and next action.

Move review earlier than the PR

Generated slide showing review moving earlier through plan, trace, diff, and evidence before the merge request checkpoint.

Evidence keeps judgment attached to the diff

Generated slide showing proof levels from low-risk formatting checks through high-risk migration dry runs, rollback, and human signoff.

The control loop decides what survives

Generated slide showing map, procedure, state, and evidence updates as the control loop for preserving useful context.

Durable context travels

Generated slide showing repo-local context traveling across fresh threads, agents, engineers, teammates, and environments.

The before and after is mundane

Generated slide comparing reactive, low-context agent work with structured durable context that makes reviews start from intent and proof.

Promotion closes the loop

Generated slide showing task-local learning distilled from work items and evidence into reusable repo guidance, so the next task starts smarter.

Thank you

Generated closing slide saying thank you and keep context close to the work, with AGENTS.md, skills, work items, and repo-local context flowing into a dot-agents folder between a human and robot feedback loop.

Looking for more resources? Visit the Coding with Agents hub →