Devtool Docs Audit - agentic-skills

2026-06-02 refresh. Alignment page path: alignment/devtool-docs-audit-agentic-skills-2026-06-02.html.

This page renders the complete refreshed audit from research/devtool-docs-audit.md, plus evidence, assumptions, alternatives, approval gates, and feedback controls.

Table of Contents

Executive Verdict

P0 None. The current docs are not missing the basic install, pack, restart, or mode concepts needed by an experienced shell user.

P1 The activation path is still too implicit. A reader can learn commands, but the docs still do not give a single first-success route from clone to visible proof.

P1 Troubleshooting remains fragmented. Restart, refresh, mode, approval-packet, drift, and validation recovery are documented, but not in one symptom-led support path.

P1 Proof/example artifacts are strong but under-used. tasks/history.md, benchmark outputs, Skills Showcase generated proof data, and devtool research artifacts demonstrate real usage, but README does not present them as the adoption proof path.

P2 Script references are better than before, but still split across multiple docs. The most common commands are covered; exact "which script answers which question" guidance is still scattered.

P2 Team rollout guidance is still research-only. Public docs cover .agents/project.json and refresh behavior, but not a concise team checklist.

P2 Stale install.sh references remain in the devtool research chain. README and skills reference now correctly use ./init.sh.

Findings

P1 - First-success route is still implicit

Claim: README and workflow docs explain the pieces, but they do not package a fastest proof path.

Evidence: README opens with ./init.sh, lists pack install commands, and explains reload/restart behavior. docs/codex-workflow.md explains $exec and tasks/todo.md. A search for first-success language in README/docs finds no dedicated section.

Inference: users must synthesize the activation sequence themselves.

Decision impact: add a README "First successful cycle" section before larger docs work.

1. Clone this repo and run ./init.sh.
2. In a target project, run scripts/pack.sh install <pack-or-skill>.
3. Restart Claude Code or start a fresh Codex session if the skill list is stale.
4. Run /pack or $pack, or scripts/pack.sh status, to confirm the project designation.
5. If tasks/todo.md is missing, run a planning skill first.
6. Run /exec or $exec and confirm tasks/history.md plus commit/push evidence.

P1 - Troubleshooting is present as fragments

Claim: the repo documents important recovery facts, but a user with a symptom still has to search across several files.

Evidence: README covers reload/restart; docs/packs.md covers doctor, refresh, stale installs, pinned installs, and update mode; docs/operating-modes.md covers approval-packet lifecycle and hybrid degraded paths; no central troubleshooting page exists.

Inference: troubleshooting content is organized by subsystem, not by user failure.

Decision impact: add a symptom-led docs/troubleshooting.md or README subsection.

P1 - Proof artifacts are strong but under-presented

Claim: the repo has credible dogfood proof, but the landing docs do not point readers to a short proof trail.

Evidence: tasks/history.md, research/devtool-*.md, docs/benchmark-results-matrix.md, and docs/skills-showcase/assets/github-proof-data.js all provide inspectable proof signals. README lacks a proof/examples section.

Inference: existing proof can be converted into adoption value with links and honest boundaries.

Decision impact: add a README "Proof and examples" block.

P2 - Script reference is scattered

Claim: common script commands are documented, but spread across README, pack docs, skills reference, operating modes, and test harness docs.

Decision impact: add docs/scripts-reference.md or expand docs/skills-reference.md with a script reference table covering install/global, project packs, mode/handoff, skill hygiene, showcase, and benchmark commands.

P2 - Team adoption checklist is still missing

Claim: team rollout concerns are known, but public docs stop short of a checklist.

Evidence: docs/packs.md explains committed project designation and generated local roots, while research docs identify checkout-path coupling, restart friction, direct-to-primary defaults, and generated-root discipline.

Decision impact: add a practical team checklist to docs/packs.md.

P2 - Existing devtool research has stale installer terminology

Claim: dogfood research artifacts still mention install.sh, but the current repo uses init.sh.

Evidence: install.sh is absent; README and skills reference use ./init.sh; rg "install.sh" research/devtool-*.md still finds references in multiple active research artifacts.

Decision impact: reconcile those artifacts or add dated historical-context notes before promoting them as examples.

Coverage By Audit Area

Quickstart Clarity

Strengths: current ./init.sh, copy-paste pack examples, reload/restart nuance.

Gaps: no named first-success route, no missing-tasks/todo.md branch, no proof endpoint beyond status.

Examples

Strengths: dogfood history, devtool research chain, benchmark/showcase proof data.

Gaps: README does not identify them as examples; some research examples use stale installer terms.

API And Script Reference

Strengths: comprehensive skill and pack catalogs, strong operating-mode reference, test harness docs.

Gaps: no single command index or failure interpretation guide.

Troubleshooting

Strengths: restart/refresh, drift mechanics, approval-packet checks, validation commands.

Gaps: no central symptom-led page; Windows/WSL and moved-checkout risks are not concise public recovery paths.

Migration Paths

Strengths: former globals mapped to packs, mode migration documented, track-latest vs pinned explained.

Gaps: team rollout guidance and stale installer terminology.

Proof Artifacts

Strengths: dogfood history, benchmark matrix, showcase data, no hidden telemetry.

Gaps: proof is not surfaced in README and should be bounded honestly.

Evidence Matrix

Claim Evidence Inference Confidence Decision Impact
No P0 docs blockerREADME includes ./init.sh, pack install, pack status/which, restart guidance; docs cover packs and modes.Experienced shell users can install and reason through setup.HighNo emergency docs rewrite needed.
First-success path is implicitNo first-success section in README/docs; workflow pieces live across README and docs/codex-workflow.md.Activation depends on user synthesis.HighAdd README first-success route.
Troubleshooting is fragmentedRecovery facts live in README, docs/packs.md, docs/operating-modes.md, and docs/skills-reference.md.Symptom-led support is slower than it needs to be.HighAdd troubleshooting table/page.
Proof is under-presentedtasks/history.md, benchmark matrix, showcase proof data, and devtool research exist; README lacks proof/examples section.Existing proof can be converted to adoption value cheaply.HighAdd proof/examples pointers.
Script reference is scatteredCommands are covered in at least five docs.A command index would reduce support overhead.Medium-highAdd script reference.
Team checklist missingMechanics documented; checklist absent; research names team friction.Team adoption still needs local tacit knowledge.Medium-highAdd team adoption checklist.
Research examples have stale installer termsinstall.sh absent; active docs use init.sh; devtool research still references install.sh.Example artifacts can confuse readers.HighReconcile or caveat stale examples.

Evidence Coverage Gate

Is the evidence sufficient to accept the audit recommendations as the next docs backlog?

Assumptions And Confidence

Assumptions Gate

How should uncertain assumptions be handled before docs remediation starts?

Alternatives Considered

  1. Full documentation rewrite: rejected. The docs are mostly accurate; the highest value is navigation and activation packaging.
  2. Only update stale install.sh references: rejected as too narrow. It fixes an accuracy issue but misses activation and support gaps.
  3. Add a separate website-first onboarding page: deferred. README and docs pages are the current adoption surface.
  4. Implement doc fixes immediately in this audit: rejected for this step. The task requested an audit refresh and alignment review before follow-up doc changes.

Recommended Backlog

  1. Add README "First successful cycle" and "Proof and examples" sections.
  2. Add docs/troubleshooting.md with symptom-led recovery steps.
  3. Add docs/scripts-reference.md or a compact script reference section in docs/skills-reference.md.
  4. Add a team adoption checklist and moved-checkout recovery note to docs/packs.md.
  5. Reconcile stale install.sh references across active research/devtool-*.md artifacts, or add dated historical-context notes before using those artifacts as examples.

Scope Gate

Which backlog scope should the next implementation step take?

Approval Gates

Artifact Destination Gate

Where should approved docs remediation write the new command reference?

Post-Approval Route Gate

What should the next agent do after this audit is approved?

Compile Section

Use feedback YAML for concerns before final approval. Use final answers when every required gate is answered.

No YAML compiled yet.