Project Config Tools

The project config file is the agent's long-term memory for this project — which tools have been confirmed, which are pending evaluation, and a running audit trail. Three local tools manage it.

File shape

{
  "version": "1.0",
  "project": {
    "name": "my-saas",
    "language": "TypeScript",
    "framework": "Next.js"
  },
  "tools": {
    "confirmed": [
      { "name": "Prisma",  "layer": "database", "version": "6.2.0" },
      { "name": "Auth.js", "layer": "auth",     "version": "5.0.0-beta.29" }
    ],
    "pending_evaluation": [
      { "name": "Drizzle", "layer": "database", "reason": "alternative to Prisma" }
    ]
  },
  "audit_log": []
}

Initialising the file

The first call to toolcairn_init creates the scaffold automatically — there's no separate "init project config" tool to call. From then on, agents read and update the file with the two tools below.

read_project_config

Parses the current config, validates the schema, and returns confirmed, pending_evaluation, and stale lists (stale = confirmed but last-seen > 30 days).

update_project_config

Mutates the config with add/remove/update operations on tools and evaluations. Every change is recorded in audit_log for traceability.

Agent pattern

1. Start of session: read_project_config to load context.
2. After a stack decision: update_project_config to record the pick.
3. Periodically: update_project_config to flip pending_evaluation entries to confirmed.