Trying To Reduce Codex Token Burn Without Adding More Noise
A measured Codex token-use note: Token Frugal helped on a small repo, but my rule-heavy setup exposed the cost of adding another instruction layer.
Personal technical journal
I write these notes first to keep track of what I learn. They are public because they may also be useful to someone else.
Start here
A local investigation into perceived Codex slowness: SQLite logs, WAL files, archived JSONL sessions, and a small macOS maintenance script.
Latest notes
A measured Codex token-use note: Token Frugal helped on a small repo, but my rule-heavy setup exposed the cost of adding another instruction layer.
A bounded Codex Dynamic Workflows lab with a local runtime, MCP server, skill, plugin packaging, real read-only campaigns, and a pause after token-heavy testing.
A local Hermes Agent installation driven by Codex: Open WebUI, OpenAI-compatible API, launchd, streaming errors, and filesystem guardrails.
An SEO lab with four Codex variants to see whether Stax helps an already-framed workflow, and whether a skill makes the result more reliable.
An MCP lab around Pipedrive: start from a concrete need, refuse the all-powerful connector, then validate real CRM workflows on disposable data.