| name | foxcode-usage-analysis |
| description | Analyze historical Codex and Claude Code sessions for real FoxCode usage. Use when evaluating how effectively agents use FoxCode, evalInBrowser, launch skills, or browser automation across past sessions. |
FoxCode Usage Analysis
Analyze past Codex and Claude Code sessions to measure real FoxCode usage.
Workflow
- Read
documents/requirements.md and documents/design.md for current FoxCode context.
- Run the bundled analyzer from the repository root:
python3 .agents/skills/foxcode-usage-analysis/scripts/analyze_foxcode_usage.py
- Treat these as the primary metrics:
actual_sessions: sessions with real status, evalInBrowser, or launch helper calls.mention_files: noisy upper bound; includes sessions where FoxCode only appears in available tools/skills.eval_sessions: sessions that actually used browser automation.reliable_flags: sessions with concrete errors (No connection, Timeout (, invalid selector, missing host permission).
- Inspect
recent_actual examples in the script output before drawing conclusions.
- Report in Russian unless the user asks otherwise.
Interpretation Rules
- Do not count a session as FoxCode usage just because
foxcode appears in the system prompt or tool catalog.
- Codex launch friction is indicated by high
launch_firefox + status counts and low evalInBrowser.
- Effective use is indicated by multiple
evalInBrowser calls after a successful connection, with few reliable flags.
text= and :has-text(...) inside evalInBrowser arguments indicate Playwright-style selector confusion; FoxCode DOM helpers expect CSS selectors.Missing host permission often means the agent navigated to a tab or URL where the extension cannot inject. Prefer api.eval(fetch(...)) from an allowed page for same-origin API checks.
Output
Return a concise report:
Scope: history locations, file counts, date range if visible.
Codex: actual sessions, eval sessions, call counts, reliable flags, main bottleneck.
Claude Code: actual sessions, eval sessions, call counts, reliable flags, main bottleneck.
Patterns: recurring inefficiencies and evidence.
Recommendations: concrete changes to FoxCode skills/docs.
If the script cannot read one IDE's history, report that path as blocked and continue with the other IDE.
