jbr870/claude-permit
1
0
Fork 0
Code Issues 17 Pull requests Projects Releases Packages Wiki Activity Actions

Add stats subcommand #1

New issue
Open
opened 2026-04-09 13:37:48 +00:00 by jbr870 · 2 comments
jbr870 commented 2026-04-09 13:37:48 +00:00
Owner
Copy link

Summary

Adds an audit-log-driven Stats section to the claude-permit report --html subcommand. Turns audit.jsonl into aggregates — session scorecards, trends, hit counts, LLM latency — rendered alongside the rules inventory built in #10.

Note: originally filed as claude-permit stats. Renamed to report --html so a single subcommand can host multiple sections (inventory, stats, future additions) without a later command rename.

Precursor

#10 — establishes the claude-permit report --html subcommand and the v1 HTML shell (header strip + rules tables, no audit data). This issue is the follow-up that layers audit-log-driven sections on top of that shell.

User Stories Served

  • US-1: Session scorecard (how many prompts saved)
  • US-2: Session summary (total calls, auto-approved, prompted, denied)
  • US-3: Trends over time (are prompts decreasing as auto-rules accumulate?)
  • US-8: LLM latency percentiles
  • US-9: Passthrough ratio (are rules stale?)
  • US-10: Rule hit counts (which rules do heavy lifting?) — needs #2 for stable rule identity
  • US-13: Tool type breakdown
  • US-14: Most frequent commands/paths
  • US-15: Session comparison

Suggested Capabilities

Rendered as sections within the HTML report from #10:

  • Overall summary (all time or last N days)
  • Per-session scorecards
  • Time-windowed aggregation (e.g. last 7 days)
  • Hit counts per rule (depends on #2)
  • Tool and command frequency tables
  • LLM latency percentiles

CLI flags to filter what the report includes:

  • --since 7d — time window
  • --session <id> — single-session focus
  • --json — machine-readable alternative to HTML (optional)

Context

See wiki: Data Capture Analysis — "Structural Issues §3: No aggregation layer"

The raw data in audit.jsonl is comprehensive. The gap is purely in consumption — no code reads the log today except the /audit skill doing manual jq-style filtering.

## Summary Adds an audit-log-driven **Stats section** to the `claude-permit report --html` subcommand. Turns `audit.jsonl` into aggregates — session scorecards, trends, hit counts, LLM latency — rendered alongside the rules inventory built in #10. > **Note:** originally filed as `claude-permit stats`. Renamed to `report --html` so a single subcommand can host multiple sections (inventory, stats, future additions) without a later command rename. ## Precursor **#10** — establishes the `claude-permit report --html` subcommand and the v1 HTML shell (header strip + rules tables, no audit data). This issue is the follow-up that layers audit-log-driven sections on top of that shell. ## User Stories Served - **US-1:** Session scorecard (how many prompts saved) - **US-2:** Session summary (total calls, auto-approved, prompted, denied) - **US-3:** Trends over time (are prompts decreasing as auto-rules accumulate?) - **US-8:** LLM latency percentiles - **US-9:** Passthrough ratio (are rules stale?) - **US-10:** Rule hit counts (which rules do heavy lifting?) — needs #2 for stable rule identity - **US-13:** Tool type breakdown - **US-14:** Most frequent commands/paths - **US-15:** Session comparison ## Suggested Capabilities Rendered as sections within the HTML report from #10: - Overall summary (all time or last N days) - Per-session scorecards - Time-windowed aggregation (e.g. last 7 days) - Hit counts per rule (depends on #2) - Tool and command frequency tables - LLM latency percentiles CLI flags to filter what the report includes: - `--since 7d` — time window - `--session <id>` — single-session focus - `--json` — machine-readable alternative to HTML (optional) ## Context See wiki: [Data Capture Analysis](https://git.wihslon.com/jbr870/claude-permit/wiki/product%2Fdata-capture-analysis) — "Structural Issues §3: No aggregation layer" The raw data in `audit.jsonl` is comprehensive. The gap is purely in consumption — no code reads the log today except the `/audit` skill doing manual jq-style filtering.
jbr870 commented 2026-04-20 19:42:10 +00:00
Author
Owner
Copy link

Related: #9 — audit subcommand for rule review-and-promote workflow. Overlaps on the audit-log parsing layer; if this ships first, #9 can build on the shared parsing code rather than duplicating it.

Related: #9 — `audit` subcommand for rule review-and-promote workflow. Overlaps on the audit-log parsing layer; if this ships first, #9 can build on the shared parsing code rather than duplicating it.
jbr870 commented 2026-04-27 13:00:34 +00:00
Author
Owner
Copy link

Sibling issue filed: #20 (friction insights). Split: this issue (#1) owns the pure descriptive stats (latency, hit counts, totals, trends — neutral data); #20 owns the opinionated friction view (promote/deny/stuck flags, suggestions). Both render as separate sections in the same HTML report.

Sibling issue filed: #20 (friction insights). Split: this issue (#1) owns the *pure descriptive* stats (latency, hit counts, totals, trends — neutral data); #20 owns the *opinionated* friction view (promote/deny/stuck flags, suggestions). Both render as separate sections in the same HTML report.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
12-auto-promote-quote-escaping
No results found.
Labels
Clear labels   
enhancement
New feature or improvement

  
observability
Insights, stats, and transparency features

  
research
Research and analysis

No labels
enhancement
observability
research
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jbr870/claude-permit#1
No description provided.