Agent Progress File Memory

Agent Progress File Memory

Agent progress file memory is the practice of keeping long-running agent state in an explicit file such as CHANGELOG.md, so future sessions inherit progress, failed attempts, checkpoints, and constraints.

Key points

  • Anthropic describes a progress file as portable long-term memory for Claude Code, functioning like lab notes across sessions [src-072].
  • A good progress file tracks current status, completed tasks, failed approaches, why they failed, accuracy tables, key checkpoints, and known limitations [src-072].
  • Recording failed approaches matters because otherwise later sessions can repeat the same dead ends [src-072].
  • The progress file complements CLAUDE.md: CLAUDE.md holds the plan and rules, while the progress file records what actually happened as the agent worked [src-072].
  • Git commits turn the progress file into coordination infrastructure, making the agent’s trajectory inspectable and recoverable from another machine or later session [src-072].

Related entities

Related concepts

Source references

  • [src-072] Siddharth Mishra-Sharma – “Long-running Claude for scientific computing” (2026-03-23)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 494 indexed pages and articles.

  1. Wiki concept Stateless Agent Memory Pattern Design pattern for agents that run in ephemeral, stateless environments (such as Claude Code cloud routines). Related by 072
  2. Wiki concept Test Oracle Driven Agents Agents whose long-running work is guided by a reference implementation, quantified objective, or test suite that lets Related by progress
  3. Insight Recommendation Systems in Production How recommendation systems become production decisioning systems through signals, ranking, constraints, feedback loops, and experimentation Readers have engaged with this next