Project Memory

Durable, evidence-backed memory for coding agents.

Project Memory gives coding agents a single, trustworthy memory for a software repository — so they stop re-learning the same facts and stop writing conflicting "memory" files.

Most repositories already contain fragments of project knowledge scattered across AGENTS.md, CLAUDE.md, .cursor/rules/, docs, and notes from previous AI tools. Project Memory treats all of that as existing knowledge to discover, verify, and reconcile — not as separate truths to leave alone.

Code tells agents what exists. Project Memory helps them remember why — and reconciles every place that already tried to write it down.

Why it matters

Without reconciled memory, every agent (and every AI tool) re-discovers the architecture, re-tries the rejected approach, and writes its own slightly different conclusion. You end up with several disagreeing "sources of truth."

With Project Memory, that knowledge is discovered, verified against the actual code, and rebuilt into one canonical memory that the next agent loads from a single trustworthy place.

Install

irm https://raw.githubusercontent.com/LoveDoLove/Project-Memory-Agent/main/install.ps1 | iex

Downloads the agent + its 8 skills into your chosen tool's global config. Pick a target from the menu (1 OpenCode, 2 Codex, 3 Claude, 4 All). Via irm | iex it defaults to all non-interactively.

Target	Skills	Agent
OpenCode	~/.config/opencode/skills/	~/.config/opencode/agents/project-memory.md
Codex	~/.agents/skills/	~/.codex/agents/project-memory.toml
Claude	~/.claude/skills/	~/.claude/agents/project-memory.md

• all writes skills to ~/.claude/skills + ~/.agents/skills (no OpenCode double-load).
• Local options: -Target all, -Verify (dry-run), -Branch dev.
• Codex: needs [features] multi_agent = true in ~/.codex/config.toml (installer prints this; never edits your config).

Use

Invoke the orchestrator in your agent:

@project-memory

It discovers existing knowledge, inspects the repository, verifies claims against real evidence, and reports what to keep, change, merge, or remove — across every origin tool, not just its own files.

Typical tasks: audit this repo, build memory for a repo that already has docs, reconcile conflicting AGENTS.md / CLAUDE.md, compound lessons after this feature.

What it does

• Discovers existing knowledge — inventories every pre-existing source with provenance.
• Verifies with evidence — claims checked against code, tests, config, build/CI, Git. Docs are not evidence.
• Classifies & resolves conflicts — one canonical type per claim; conflicts settled by repository evidence, not by file age.
• Compounds learning — durable Solutions and Lessons instead of more documents.
• Removes obsolete knowledge — deleted or marked deprecated/superseded, regardless of origin.
• Single source of truth — one primary home per concept; everything else references it.
• Self-audits — a quality bar (Memory Health) and a Self-Audit directive let the agent run its own pipeline on its own repo to catch drift.

Skills

Specialized Skills instead of one huge prompt:

Skill	Responsibility
knowledge-discovery	Inventory every pre-existing knowledge source, with provenance
repository-audit	Gather repository evidence
knowledge-classification	Classify and resolve cross-source conflicts
knowledge-compounding	Turn experience into reusable memory
memory-architecture	Design hierarchy, navigation, reconstruction
obsolete-knowledge	Handle stale or superseded knowledge
memory-edit	Apply approved documentation changes
memory-verification	Final consistency and quality gate

The orchestrator loads these progressively — you rarely invoke one directly. It can also delegate to codebase-memory (read-only code graph) and cavecrew-builder (bounded edits).

How memory is organized

AGENTS.md is the single entry point; everything else is referenced, not duplicated.

AGENTS.md
  → docs/<domain>/README.md      (domain index)
    → docs/<domain>/<topic>.md   (one concept per file)

Domains: architecture/, decisions/, lessons/, workflows/, constraints/, reference/, history/. This is a pattern, not a mandatory scaffold — only create what holds verified knowledge.

Testing

Windows PowerShell 5.1, Pester 3.4.0:

Invoke-Pester ./install.tests.ps1

7 tests cover installer targets, the no-double-load rule, and a guardrail that keeps the 8 skills, their manifest, and both agent files in sync.

Status

Core agent + full memory lifecycle are complete, and the agent self-audits its own repository. See GitHub Issues for planned work (stale-reference detection, broader platform support, cross-project patterns).

License

MIT — see LICENSE.

---