Codex Weekly Usage — July 13–19, 2026
A concentrated Hyphenomenon delivery week: 29 de-duplicated Codex sessions across 57 rollout logs, an early-week push, a two-day evidence gap, and a weekend return with verified delivery to two repositories.
- Period
- Jul 13, 2026 00:00 EDT to Jul 20, 2026 00:00 EDT
- Generated
- Jul 21, 2026
- Validation
- Passed
- Analyzer
- codex-usage-report-v5
De-duplicated by session_meta.session_id; 57 active rollout log files are tracked separately.
Verified from terminal outcomes; operations are not independent feature units.
Unique paths in successful patch-application records.
codex-skills and hyphenomenon
Executive Summary
The week formed three distinct phases: 18 de-duplicated Codex sessions started Monday through Wednesday, no session starts or Chronicle summaries were recorded Thursday or Friday, and 11 sessions marked a weekend return. Hyphenomenon was the center of gravity, accounting for 10,798 of 12,251 tool calls (88.1%), while the top three rollout logs carried 51.78% of recorded tool activity. Terminal outcomes verify 90 successful commit operations, 66 successful content-push operations, and delivery output naming Hyphenomenon and codex-skills; those operation counts are evidence of execution, not independent feature units or elapsed effort.
Activity By Day And Clock Hour
Unique sessions by local start day, followed by tool calls aggregated by local clock hour across the full window.
Unique Sessions By Day
De-duplicated session_meta.session_id values assigned to their recorded local start day.
Total: 12,251 tool calls. “c” marks successful commit operations in that hour.
Activity Concentration
Session starts peaked on Wednesday (9), with Monday contributing 7, Tuesday 2, Saturday 6, and Sunday 5. The 1 PM and 7 PM local clock-hour buckets together contained 4,315 of 12,251 tool calls (35.22%) across the week. The Thursday-Friday gap appears in both Codex session starts and Chronicle summaries, but it proves only an evidence gap in these sources, not inactivity.
Where The Work Landed
Tool volume and patch churn answer different questions, so they are shown side by side instead of collapsed into one repo count.
Tool Calls By Repository
Working repository resolved from tool workdir or turn context.
Patch Churn By Repository
Added plus removed lines in patch inputs. Diagnostic only, not a productivity score.
Delivery Extended Beyond One Repository
Hyphenomenon dominated resolved tool activity, but it was not the whole operating surface. Another 1,352 tool calls (11%) occurred outside resolved git metadata, and smaller command-level traces resolved to shopify-headless, bpe-marketing-lp, codex-skills, and shopify-ai-product-optimizer. Only Hyphenomenon and codex-skills appear in successful push output, so the other repository labels are activity evidence rather than delivery claims.
Output Was Concentrated
A small number of rollouts carried most patch churn, while file touches skewed toward documentation and governance work.
Patch Churn By Rollout
Top-level | Outside git metadata
Top-level | hyphenomenon
Top-level | hyphenomenon
Top-level | hyphenomenon
Top-level | hyphenomenon
Top-level | Outside git metadata
File Touch Mix
Output Shape
Repeated file-touch records leaned toward documentation and governance (1,238) over application and pipeline code (654), with 406 unique patched paths across the week. That pattern fits a week centered on intake, report infrastructure, review packets, and delivery governance, while still carrying implementation work. Touch and patch counts can repeat paths and reflect analyzer mechanics; they describe output composition, not volume produced or productivity.
Verified Delivery Ledger
Outcomes come from terminal return records. Intent-only command counts remain in the technical appendix.
Command Outcomes
Repositories In Successful Push Output
Verification Was Part Of The Work
Terminal outcomes record 90 successful commit operations, 66 successful content pushes, 61 successful pull-request operations, and 6 successful branch-cleanup pushes. The evidence also retains 12 content pushes with unknown outcomes, 3 pending pushes, 28 pull-request operations with unknown outcomes, 9 pending pull-request operations, and 4 failures. Successful operations are verified outcomes, but the unresolved tail prevents a claim that every attempted delivery completed cleanly.
Agent Stack Used
Observed invocations only. Available tools are excluded; skill rows distinguish an explicit request from an observed SKILL.md load.
Apps and Connectors
Completed connector calls recorded by the runtime.
Tool Providers
Completed MCP provider calls, grouped by server namespace.
Skills
Explicitly requested or loaded skills; application is not inferred.
Agent Stack Evidence
Completed connector calls were led by Vercel (449) and Figma (250), followed by Google Drive (87), Shopify (49), Gmail (44), and GitHub (1). Completed MCP provider calls were concentrated in the Node tool runtime (1,263) and Playwright browser (320), with 4 Shopify Dev MCP calls. Skill rows include only explicit requests or literal SKILL.md loads; gmail, frontend-app-builder, control-in-app-browser, hyphenomenon-project-intake, and playwright were the most frequently observed labels, but a load does not prove the skill was applied.
Chronicle Context Across The Week
Chronicle supplies contextual coverage across desktop work. Lane counts overlap and should not be summed as duration.
Work Lane Mentions
Evidence Coverage
Chronicle Context
Chronicle contributes 450 overlapping ten-minute summaries, 3,905 citation occurrences, and 1,092 unique cited minutes across 18 observed clock hours. Planning and knowledge work appeared in 400 summaries, software and repo delivery in 345, print and artwork production in 323, storefront and catalog work in 316, and wholesale or shipping in 145. The overlap shows a mixed operating week beyond repository work; the lane totals must not be summed as duration, and the missing Thursday-Friday summaries remain an explicit coverage limitation.
Technical Telemetry
Raw tool, event, workspace, and patch diagnostics used to audit the analyzer. These counts are intentionally secondary to outcomes.
Tool Surfaces
Function Calls
Custom Tool Calls
Response Events
Patch Applications
Output Records
Session And Log Audit
Anomalies And Uncertainty
The source contains 248 raw session_meta records but only 29 unique session_meta.session_id values across 57 active rollout log files; the report uses the de-duplicated session count and exposes the other two figures only as audit fields. Tool, patch, and file-touch telemetry can repeat across retained rollout histories, while delivery operations are keyed to terminal call outcomes. Truncated outputs (645), long-running handoffs (205), and non-zero terminal exits (139) are retained as diagnostic signals and are not treated as proof of failed work, duration, or productivity.