production-review
Skill14 starsChanged 4 months agoMITView on GitHub
Run a full production-readiness review of santree-app (multi-agent fan-out, adversarial verification, PRODUCTION_REVIEW.md with checkbox tracking), or burn down an existing PRODUCTION_REVIEW.md via the disjoint-file wave pattern. Use when asked for a full/production/pre-release review of the codebase, or to fix the findings of one. Pass `fix` as args to go straight to fix mode.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- ragna-ai/ragna-studiotheir own version, first seen Nov 8, 2026
- frankxai/creator-skillstheir own version, first seen Oct 17, 2026
- srednoff888-art/srednoff-os-for-claudetheir own version, first seen Jul 1, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- frankxai/creator-skills★ 6same name
- ragna-ai/ragna-studio★ 1same name
- srednoff888-art/srednoff-os-for-claude★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- new-commandSkill
- production-reviewSkill
- verifySkill
- new-commandSkill
- verifySkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Use the Workflow tool for the fan-out (it's resumable via resumeFromRunId — a previous run of this review died mid-dispatch and lost …
- Distinctive: line 40
Phase 2 — adversarial verify: every High (and any low-Confidence) finding gets an independent skeptic agent prompted to REFUTE it against …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Production-readiness review
Two modes. Review mode produces/refreshes PRODUCTION_REVIEW.md. Fix mode
burns down an existing PRODUCTION_REVIEW.md. The user prefers these as separate
sessions — don't slide from one into the other unasked.
Both modes: read AGENTS.md and COMPLIANCE.md first. If PRODUCTION_REVIEW.md
already exists, read it — its own checkboxes are the source of truth for what's
done; never trust remembered "N of M fixed" counts, they go stale.
Review mode
Use the Workflow tool for the fan-out (it's resumable via resumeFromRunId —
a previous run of this review died mid-dispatch and lost everything; a workflow
run would not have).
Phase 1 — finders, one agent per disjoint slice, each returning structured
findings. Slices: core crates (crates/core, crates/pty); Rust backend split
into 2–3 disjoint file groups (commands.rs+lib.rs / linear.rs+github.rs+
reviews.rs / db.rs+repo.rs+settings.rs+terminal.rs+git.rs+migrations);
frontend data layer (queries.ts, main.tsx, state/, bindings.ts usage);
each feature view (features/*); shared UI (components/, theme/); ops
(CI, bundling, logging, updater). Every view is real — nothing is mocked, so
nothing is out of scope on "still mocked" grounds.
Categories per finder: correctness; React/TanStack idioms; Rust/tokio/Tauri idioms; security (the AGENTS.md invariants: IPC path/id/branch validation, parse-at-sink host matching, keychain-only secrets); COMPLIANCE.md drift; perf; architecture; readability; UX responsiveness (optimistic updates); ops readiness; testing gaps; deps; a11y.
Per-finding format: Title / Location (file:line) / Category / Severity /
Effort / Confidence / Problem / Why it matters / Suggested fix.
Phase 2 — adversarial verify: every High (and any low-Confidence) finding gets an independent skeptic agent prompted to REFUTE it against the actual code. Drop refuted findings.
Phase 3 — synthesize: dedupe, then triage with this project's philosophy
(zero users, runs locally): bucket findings as (1) Do now — bites daily
local use, especially silent data loss/repo corruption; (2) Before
distributing — signing, updater, release pipeline; (3) Security,
threat-model-dependent. Lead with tier 1; severity labels alone are not the
ordering. Respect existing ⏸ deferred markers (e.g. signing/notarization) —
don't re-raise parked work.
Write PRODUCTION_REVIEW.md with checkbox tracking: every findings table
leads with a Done column ([ ] per row); quick wins, roadmap, and readiness
checklist are real - [ ] task lists; genuinely-clean categories are
pre-checked [x]; a short "how to use this doc" note at the top states the
rule: only flip [ ] → [x] when the fix is real and verified.
Fix mode
Work order: quick wins → phase 1 → 2 → 3, higher severity first within a phase.
Wave pattern (proven across 5+ waves, ~150 findings, zero surviving regressions):
- Group findings into disjoint-file batches — two concurrent agents never
touch the same file. Give each agent the finding's full tracker text, not a
summary. Big structural findings touching shared state (
model.tsx,queries.ts,domain.rs,linear.rs) are done serially by the main session, not parallelized. - Dispatch the whole wave at once; don't verify agents one-by-one.
- One consolidated verification pass after the wave lands — the /verify skill's static gates, in its order. Only then spot-check each finding's fix is actually in the diff.
- Never chase mid-wave failures. Agents report "pre-existing" compile/lint errors that are just another agent's edit mid-flight; they resolve when the wave lands. Investigate only failures still present in the consolidated pass.
- Fixes must be real: open the cited code → confirm it still reproduces →
fix → verify the behavior changed → only then tick. A non-reproducing
finding is a valid outcome: leave
[ ]with a note, never force a tick. - Only the orchestrating session edits
PRODUCTION_REVIEW.md— subagents never touch it. - Autonomy: internal/security/perf fixes — just do them. User-visible changes (new UI, copy, defaults) — batch into one question, don't interrupt per item.
- High-risk changes (wire formats, PTY, migrations) get manual review beyond the agent's report: read the diff, exercise it live.
- At the end, clean up stray agent worktrees/stashes after confirming their contents are superseded in the main tree. Leave committing to the user.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

