Prompt History Backfill — Full Audit
Generated 2026-06-04 • Mode: report-only • Repo: agentic-skills
Scope & Sources
Repository: /Users/georgele/projects/tools/agentic-skills
Date range: All history (no date filter applied)
Mode: Report-only (no files written)
Sources Read
| Source | Status | Size |
|---|---|---|
| ~/.claude/history.jsonl | OK | 9,964 lines |
| ~/.codex/history.jsonl | OK | 6,543 lines |
| ~/.codex/sessions/**/*.jsonl | OK | 2,676 session files |
Sources missing: None
Summary Statistics
Invocations by Source
| Source | Unique High-Confidence | Distinct Skills | Date Range |
|---|---|---|---|
| Claude Code | 54 (115 skill×session pairs) | 18 | 2026-04-07 – 2026-06-04 |
| Codex | 599 (604 skill×session pairs) | 20 | 2026-04-07 – 2026-06-03 |
Existing prompt files cover May 30 – Jun 4 only. April – late May invocations are almost entirely uncaptured.
Skills Missing Prompt Directories
These skills were invoked in conversation history but have no prompts/ directory at all. Running --apply would create directories and backfill prompt files for high/medium confidence matches.
| Skill | Claude Invocations | Codex Invocations | Total | Priority |
|---|---|---|---|---|
| run | 10 | 75 | 85 | High |
| targeted-skill-builder | 3 | 56 | 59 | High |
| benchmark-agent-review | 1 | 32 | 33 | High |
| provision-agentic-config | 2 | 0 | 2 | Medium |
| skill-interview | 2 | 1 | 3 | Medium |
| code-review | 1 | 0 | 1 | Medium |
| youtube-audit | 1 | 0 | 1 | Medium |
| review | 0 | 3 | 3 | Medium |
| idea-scope-brief | 0 | 1 | 1 | Medium |
| expert-review | 0 | 1 | 1 | Medium |
| verify | 0 | 1 | 1 | Medium |
| simplify | 0 | 1 | 1 | Medium |
| schedule | 0 | 1 | 1 | Medium |
Backfill would create: 13 new prompts/ directories and an estimated 192 prompt files for these skills.
Gap Analysis — Existing Directories
These skills already have prompt directories but are missing earlier invocations (mostly April – late May). The "gap" column shows uncaptured invocations.
| Skill | Existing Files | Total Invocations | Gap | Coverage |
|---|---|---|---|---|
| ship | 4 | ~307 | ~303 | 1% |
| benchmark-test-skill | 2 | ~85 | ~83 | 2% |
| session-triage | 3 | ~37 | ~34 | 8% |
| analyze-sessions | 2 | ~34 | ~32 | 6% |
| ship-end | 5 | ~32 | ~27 | 16% |
| exec | 24 | ~28 | ~4 | 86% |
| pack | 8 | ~18 | ~10 | 44% |
| sync | 2 | ~15 | ~13 | 13% |
| investigate | 14 | ~15 | ~1 | 93% |
| skills | 1 | ~87 | ~86 | 1% |
| init-agentic-skills | 2 | ~2 | 0 | 100% |
| devtool-docs-audit | 1 | ~1 | 0 | 100% |
| commit-and-push-by-feature | 1 | ~1 | 0 | 100% |
| skill-creator | 3 | ~3 | 0 | 100% |
| positioning | 1 | ~1 | 0 | 100% |
Note on bare invocations: Many /ship, /run, /sync, /skills, and /ship-end invocations are bare commands with no substantive prompt text. Backfilling bare /ship (prompt = just "/ship") may add volume without value. The --apply run would still capture them per the convention, but consider filtering to substantive prompts only.
Representative Uncaptured Prompts
High-value uncaptured prompts with substantive user context (not bare commands):
| Skill | Date | Source | Prompt (excerpt) | Confidence |
|---|---|---|---|---|
| analyze-sessions | 2026-04-11 | Claude | "when we are doing the run and ship loops, I notice that there are pre-existing issues identified after a run..." | High |
| analyze-sessions | 2026-04-28 | Claude | "determine if based on conversation history and past plan-interviews... if there are any issues from the par..." | High |
| analyze-sessions | 2026-04-29 | Claude | "analyze the recent bismarck-v5 attempt to parallel work using codex..." | High |
| analyze-sessions | 2026-04-30 | Claude | "how are the codex sessions running gpt 5.5 low vs gpt 5.5 xhigh doing?..." | High |
| analyze-sessions | 2026-05-01 | Claude | "we have some holes in our skills where recommendations for next steps are broken..." | High |
| analyze-sessions | 2026-05-07 | Claude | "audit and compile a report on how we've been using the ux-variations skill and ui-interview..." | High |
| investigate | 2026-04-19 | Claude | "Do you agree with these recommendations? Do not implement anything..." | High |
| investigate | 2026-05-27 | Claude | "evaluate our skills to ensure we can properly support branching research..." | High |
| investigate | 2026-06-03 | Claude | "audit all of the skills to determine if there any language inconsistency..." | High |
| investigate | 2026-06-03 | Claude | "evaluate our skills list and determine if we should have a skill called repo glossary..." | High |
| skill-interview | 2026-05-20 | Claude | "we need a skill to help migrate an existing project to our new workflow..." | High |
| skill-interview | 2026-05-20 | Claude | "create a quiz me skill which will interrogate the user as to their understanding..." | High |
| targeted-skill-builder | 2026-05-20 | Claude | "desk-flip benchmark artifact-reference false-negative" | High |
| benchmark-test-skill | 2026-05-20 | Claude | "desk-flip" | High |
| benchmark-agent-review | 2026-05-20 | Claude | "desk-flip" | High |
Coverage Checkpoint Gate
Gate: Coverage Checkpoint
The audit found ~700 high-confidence invocations across 28 distinct skills, but only 67 prompt files exist covering ~5 days of a 2-month history window. The bulk of uncaptured invocations are bare commands (/ship, /run, /sync) with minimal prompt text.
Q1: Should bare command invocations (prompt text ≤ 20 chars, e.g. just "/ship") be backfilled?
Gate: Backfill Scope
13 skills have zero prompt history. 10 skills have directories but significant gaps.
Q2: Which skills should be prioritized for backfill?
Approval Gate
Gate: Proceed to Apply
Q3: Ready to run /prompt-history-backfill --apply with the selected scope?
Verification Notes
- All candidate prompts extracted from visible user-facing fields only (display, text, leaf).
- No hidden system/developer instructions, tool output, or assistant messages included.
- No prompt files were created or modified (report-only mode).
- No likely secrets detected in sampled candidate prompts.
- All proposed writes would go under
prompts/<skill-slug>/only.
Compile Section
3 required questions unanswered
Copied!