audit-dispatch-skills-fix
Triage and explain audit defects and opportunities, then deliver coherent batches through Dispatch planning and review with direct implementation.
Skill by Gyunikuchan
Chalk Labs
Need help deploying audit-dispatch-skills-fix?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/gyunikuchan-audit-dispatch-skills-fix
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: security
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Node.js package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/audit-dispatch-skills-fix
What's inside
- Skill name
audit-dispatch-skills-fix- Description
- Triage and explain audit defects and opportunities, then deliver coherent batches through Dispatch planning and review with direct implementation.
- License
- MIT
- Metadata
- internal: true
- Instructions
- 74 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (3)
- SKILL.md
- triage-and-state.md · references/
- status.mjs · scripts/
SKILL.md (first lines)
# Audit Dispatch Skills Fix Turn an `audit-dispatch-skills` report into verified changes. Recheck each claim against current source and governing intent. Keep **A-items** as defect claims and **O-items** as improvement hypotheses; selection never proves a benefit. The report owns remediation state: per-item `Status`, `Triage`, and `Execution` lines, retained execution history, and separate counts in sections 3 and 4. Supporting plans, reviews and test logs provide evidence. `<skill>` means the directory holding this file. Read [triage and state](references/triage-and-state.md) before recording metadata, handling a rejected command, or recovering an interrupted batch. Fix each batch through **Dispatch plan → Dispatch plan review → direct implementation → Dispatch code review**. Do not start Dispatch implement: a batch can modify that same implementation flow while it is active. Read the repository's `AGENTS.md` and the shipped `dispatch` contract and selected verb guides before using Dispatch. If Dispatch is unavailable, report the missing dependency and preserve the triage state. ## 1. Open and bound ```text node <skill>/scripts/status.mjs init --run <run-or-report-path> node <skill>/scripts/status.mjs list --run <run-or-report-path> --full ``` Use the user's report; without one, `init` selects the newest `.scratch/audits/<run>-audit.md`. Keep that identity on **every command**. Init preserves statuses, notes and metadata; new defects default to `open`, opportunities to `decision`. A zero-defect report can still have opportunities. Confirm the objective, selected item IDs, constraints and verification scope from the request. Triage all items unless the user narrows scope. Opportunity implementation requires explicit selection, which can come from the invocation or an earlier user instruction selecting named items or all opportunities; retain that quote instead of asking again. Unselected opportunities still receive an assessment and recommendation. **Done when:** the report identity and scope are known, counts match both sections, and existing batch records have been checked for resume. ## 2. Triage and explain For each in-scope item, inspect current cited source, producer/consumer paths, relevant tests, counterevidence, and the governing outcome. A prior `dispatched` note requires inspecting the current tree and retained artifacts before repeat work. Reproductions must be bounded and relevant. - **Defect:** confirm the actual failure and impact; reject a contradicted claim with the source or commit evidence. - **Opportunity:** assess its hypothesis, benefit, cost, trade-offs and simplest useful change. Preserve `unmeasured` claims; define how the hypothesis will be checked. Use `decision` while selection or an intent choice remains unresolved; defer a declined or unsupported improvement with a reason. - **Both:** resolve overlapping root causes, identify prerequisites, and define scoped paths, acceptance checks, priority and a coherent group. A dependency is settled only by verified `fixed` status; document duplicates in the recommendation rather than implement the same change twice. Record the ruling and explanation with `triage <id> --from <json-file>` using the reference schema. Explain to the user, by ID, what is real, why it matters, what you recommend, and what evidence supports the ruling. Ask unresolved intent/trade-off questions together, with your recommendation; continue independent work while answers are pending. Update triage from actual answers. **Done when:** every in-scope item has a recorded ruling, evidence, impact and recommendation; accepted items also have selection where required, bounded paths, dependencies and verification. ## 3. Form a coherent batch ```text node <skill>/scripts/status.mjs batch --run <run-or-report-path> ``` The command selects one triaged group, with prerequisites first. Defects rank by severity; selected opportunities follow by declared triage priority. Size is a ceiling (default 4, `--size` up to 25), not a target. Group items by shared behavior, dependencies, paths and risk; same-file proximity alone does not establish coherence. Combine a selected opportunity with a defect only when they share one change and compatible verification. Review the printed group against intent. Resolve cycles, oversized groups, missing prerequisites or inadequate scope by re-triaging before planning; report blocked items and proceed with independent ready groups. A `Resume` result sends you to the recorded phase, not a fresh batch. Explain the IDs, shared outcome, ordering, risk and completion checks. Choose Dispatch level from its current criteria; do not use a fixed severity-to-level table. **Done when:** the batch has one checkable outcome, complete dependencies and authorized scope, with no unresolved decision included. ## 4. Plan and review Invoke `/dispatch <level> plan:` with the batch IDs and report path, triage rationale, selected opportunity quotes, scope, dependencies and acceptance checks. Follow Dispatch's run loop and retain the returned plan and run identity. Its plan flow includes plan review: inspect the completion evidence and final review resolutions; this satisfies the plan-review phase. If the plan lacks a completed review, run `/dispatch <level> review: <plan-path>` and settle it before edits. Record `planned`, then `plan-reviewed` checkpoints through `progress <ids> --from <json-file>`, linking the plan and its review evidence. Inspect review findings against source and user decisions. Revised scope must remain within authorization; material intent choices go to the user. An invalid batch can be abandoned with its evidence preserved, then re-triaged. **Done when:** the exact plan is settled and reviewed, findings are resolved or explicitly bounded, scope/criteria are concrete, and both checkpoints are in the report. Disclose `fixedUnreviewed` or missing provider coverage; require further review when it leaves acceptance uncertain. ## 5. Implement directly and review code Implement the settled plan yourself in the host's normal editing loop. Use its scoped paths and task order. Run its RED checks before behavior changes or record justified exceptions; run plan Verify commands and required repository gates after edits. Preserve unrelated work. Stop before an unresolved scope or intent change; revise and review the plan when its governing outcome changes. Save criterion-by-criterion evidence and test results, then record `implemented`. Invoke `/dispatch <level> review:` on the exact batch changes with the settled plan, intent and deviations as context. Isolate the target from unrelated changes. Adjudicate claims against source; apply accepted in-scope fixes directly and rerun affected checks. Start another code review when material fixes or coverage gaps require it. Review completion alone does not prove tests passed. **Done when:** every plan criterion has post-change evidence, required gates pass, code review is complete with findings settled, and `code-reviewed` links the review plus current verification evidence. Unresolved failures remain open. ## 6. Record, resume and hand off Set each verified member to `fixed` with a concrete outcome and evidence note. The helper requires accepted triage and a `code-reviewed` checkpoint. `fixed` on an opportunity means the selected change and checks completed; report measured benefit separately. After interruption, read `Execution` and inspect its linked artifacts and current tree. Resume the same Dispatch plan/review run where active; after `implemented`, continue code review rather than repeat implementation. A write or test result lost before checkpointing must be inspected before retry. Abandon a stale batch with a reason before regrouping; retain its history and artifacts. Loop while ready work remains. Hand off with the report and plan/review links, separate status counts for defects and opportunities, delivered behavior, verification and review coverage, unresolved decisions with recommendations, deferred items with what would settle them, and contradicted claims with evidence. Include the repository's edit-task handoff fields when files changed. **Done when:** every in-scope item is fixed, rejected with evidence, deferred with a reason, or explicitly awaiting a named decision; no active batch or failed required gate is described as complete.
From Gyunikuchan/dispatch-skills, MIT.
More from Gyunikuchan/dispatch-skills
- 0.567score
audit-dispatch-skills
Gyunikuchan
Report-only static audit of Dispatch through scenario walkthroughs of each verb, shared handoff review, and bounded provider probes.
SkillAFreshness: AClaude Code +9 - 0.567score
optimise-tests
Gyunikuchan
Measure this repository's test suite, refactor it unit by unit with parallel scoped writers, and prove no unapproved coverage loss with a runtime and…
SkillAFreshness: AClaude Code +9 - 0.565score
publish-git
Gyunikuchan
Prepare and publish a GitHub release when the user requests release publication, including version selection, changelog, integration, and recovery.
SkillAFreshness: AClaude Code +9 - 0.564score
dispatch
Gyunikuchan
Multi-agent planning, review, implementation.
SkillAFreshness: AClaude Code +9 - 0.558score
dispatch-code-review
Gyunikuchan
Alias for `/dispatch review` of code.
SkillAFreshness: AClaude Code +9 - 0.558score
dispatch-implement
Gyunikuchan
Alias for `/dispatch implement`.
SkillAFreshness: AClaude Code +9 - 0.558score
dispatch-plan
Gyunikuchan
Alias for `/dispatch plan`.
SkillAFreshness: AClaude Code +9 - 0.539score
grilling
Gyunikuchan
Grill the user relentlessly about a plan, decision, or idea.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 7, 2026 (24 days ago)
- Last release
- Oct 21, 2026 (10 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/Gyunikuchan/dispatch-skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Gyunikuchan
- 0.567score
audit-dispatch-skills
Gyunikuchan
Report-only static audit of Dispatch through scenario walkthroughs of each verb, shared handoff review, and bounded provider probes.
SkillAFreshness: AClaude Code +9 - 0.567score
optimise-tests
Gyunikuchan
Measure this repository's test suite, refactor it unit by unit with parallel scoped writers, and prove no unapproved coverage loss with a runtime and…
SkillAFreshness: AClaude Code +9 - 0.565score
publish-git
Gyunikuchan
Prepare and publish a GitHub release when the user requests release publication, including version selection, changelog, integration, and recovery.
SkillAFreshness: AClaude Code +9 - 0.564score
dispatch
Gyunikuchan
Multi-agent planning, review, implementation.
SkillAFreshness: AClaude Code +9 - 0.558score
dispatch-code-review
Gyunikuchan
Alias for `/dispatch review` of code.
SkillAFreshness: AClaude Code +9 - 0.558score
dispatch-implement
Gyunikuchan
Alias for `/dispatch implement`.
SkillAFreshness: AClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get audit-dispatch-skills-fix running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
audit-dispatch-skills-fix. Agent Mart. https://agentmart.dev/skills/gyunikuchan-audit-dispatch-skills-fix (last updated 2026-10-24)