Codex Hygiene v0.3.0

Codex Hygiene v0.3.0 expands the local activity review into a system-wide, privacy-preserving review of Codex usage.

Changed

• System-wide Codex activity reviews now run rollout enrichment by default at every candidate size. The 512 MiB automatic cutoff and guarded rerun workflow were removed.
• Reviews now report all installed plugins, MCP state, Apps availability, project-stanza counts, model and token shares, and reasoning effort by model.
• Plugin weight now covers matched calls, runtime, serialized output, skill reads, and inventoried tools. Exact per-plugin context and billing tokens remain explicitly unknown.
• Reports now end with an optional prompt for identifying evidence-based token-efficiency opportunities.

Added

• A versionless GitHub social-preview image.

The review remains read-only and keeps prompts, responses, thread titles and identifiers, commands, tool results, full paths, configuration values, and secrets private.

Codex Hygiene v0.2.0

Changelog

v0.2.0 - 2026-07-17

Added

• Privacy-preserving activity reviews for explicit 1-90 day lookbacks, with Markdown or JSON output and comparison to the preceding period.
• Retention-aware source coverage that keeps current observations visible while suppressing comparisons when the required log windows are incomplete.
• Codex activity review findings label confidence across observed models, reasoning levels, tools, skills, plugins, compactions, subagents, automations, and local cumulative token changes.
• Codex activity review enrichment uses a rollout-size guard for task timing, relative serialized tool-output weight, verification-command counts, explicit compactions, and observed SKILL.md reads.
• An activity-review interpretation reference and fixture coverage for attribution, retention gaps, rollout guards, compact-mode fallback, and privacy exclusions.

Compatibility

• Activity review compatibility preserves the original compact measure_codex_context.sh workflow as the default when no lookback period or retrospective is requested. Its commands and output are unchanged.
• The activity review uses Python 3 and the standard library. Codex telemetry schemas, retained history, cache layouts, and CLI output remain version-dependent diagnostic inputs.

Privacy and limitations

• Activity review privacy excludes prompts, responses, thread titles and identifiers, commands, tool results, full paths, configuration values, and secrets.
• Activity review limitations treat token changes as local cumulative telemetry rather than billing totals. Tool and task runtimes can overlap and do not measure model inference time or work quality.
• Activity review enrichment retains a 512 MiB default guard. When candidates exceed it, the report provides an exact --max-auto-rollout-mib N guarded rerun and keeps deep-only values labeled as not measured.

Codex Hygiene v0.1.0

Codex Hygiene v0.1.0 is a read-only-first skill for auditing and tuning Codex Desktop context and tool surfaces.

Highlights:

• Measures recent MCP tool-list and snapshot-cache telemetry with per-thread cumulative token deltas.
• Separates plugin and MCP enablement from model-visible tools, and cached app inventory from current enabled state.
• Provides current per-app and global Apps controls, project-local MCP guidance, and careful long-running goal context guidance.
• Keeps output compact and avoids full logs, configs, schemas, secrets, and missing-path lists.
• Adds portability checks and a deterministic integration test for the measurement script.

Compatibility:

• Designed for macOS and Unix-like environments with Bash, sqlite3, Perl, awk, and sort.
• Uses jq and the Codex CLI when available for additional summaries.
• Local telemetry databases and caches are version-dependent diagnostic inputs, not billing records.