create-verification
Bootstrap or extend a repository's project-local verification - the VERIFY.md contract, a `verify` mise task built from checks the project already declares, a…
Skill by douglasjarquin
Chalk Labs
Need help deploying create-verification?
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/douglasjarquin-create-verification
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: documents, productivity
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
- Python package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/create-verification
What's inside
- Skill name
create-verification- Description
- Bootstrap or extend a repository's project-local verification - the VERIFY.md contract, a `verify` mise task built from checks the project already declares, a feature-map index with seed feature files, and the vendored verify/maintain skills - then prove the generated instructions by running them once. Works in any clone without sum or Herdr.
- License
- MIT
- Instructions
- 103 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (3)
- SKILL.md
- ATTRIBUTION.md · references/
- verify_scaffold.py · scripts/
SKILL.md (first lines)
# Create verification You are writing for the next agent, who will read the output cold, mid-task, without having seen this repository. Everything you generate must be something that agent can run. A contract or map that was never executed is a draft, not completed onboarding. This skill is portable: it needs Git, mise, Python 3.11+, and the two sibling skills `verify` and `maintain-verification` beside it. No sum installation, Herdr session, `.sum` directory, or absolute path outside the target repository is involved. Run every command below from the repository you are standardizing (`--root` accepts any directory inside it). ## 1. Inspect before asking Run the scaffold in its default inspect mode: ```sh python3 <path-to-this-skill>/scripts/verify_scaffold.py --root . --json ``` It reports without writing or running anything. The report lists guidance files (README, AGENTS, ARCHITECTURE, CONTRIBUTING, CODEOWNERS, CI workflow directories), the repository's own mise tasks against inherited ones, package scripts, make/just targets, declared containers, readiness endpoints and ports the repository already mentions, CLI entrypoints, HTTP routes found in source, test locations, and which driver capabilities are actually installed (`python3`, `node`, `curl`, `docker`, `chrome-devtools-axi`, `playwright`, `tmux`). It also lists a `readme_map` of task to owner to example to check (`kind` is `observed` or `gap`), and `verify_cycle` when a `verify` task would call `mise run verify` again. File existence is observation, not enforcement. Read the guidance files it lists yourself. Ask the user only what the repository cannot tell you: which surface is primary when several exist, and where a fixture or seed data comes from when nothing documents it. Never invent a selector, command, flag, port, or credential; if the repository does not state it, the map says `manual` or carries a placeholder. If the checkout does not build or its existing checks do not pass as-is, report the exact cause and stop at a draft. Do not write a recipe around a broken base, and never add a `verify` task that merely prints success. ## 2. Generate ```sh python3 <path-to-this-skill>/scripts/verify_scaffold.py --root . --write ``` The scaffold creates only what is missing and never overwrites: - `VERIFY.md` in the shape the `verify` runner validates (`entrypoint = "mise run verify"`, `feature_maps`, `artifacts`, `[requires]`, and the Setup / Readiness / Automated checks / Scenarios / Isolation / Artifacts / Teardown / Policy sections). An existing `VERIFY.md` that declares another `feature_maps` location keeps it; new feature files go beside that index. - A `verify` task appended to `mise.toml` when the repository has none, composed from the check tasks, package scripts, make targets, or test directory the inspection found. With nothing to reuse it writes no task and reports the problem instead. - `docs/features/README.md` (or the declared index) labelled `Inventory: incomplete`, plus one file per seed feature: the top three to five real user-facing entrypoints it found, each with a stable ID, an Entry points section, a Scenarios table whose rows the runner reads, Driving it, Expected states and side effects, and Gotchas and manual gaps. Everything it could not observe is an explicit `TODO(verify)` placeholder. - `.agents/skills/verify/` and `.agents/skills/maintain-verification/` vendored byte for byte, with thin `.claude/skills/<name>` symlinks. Substantive logic stays in the `.agents/skills` copy; the aliases only make native discovery work. - `.artifacts/` added to `.gitignore`. An existing file is never rewritten. One without `TODO(verify)` placeholders is yours and is reported `kept`. A draft that still has placeholders but that the inspection would now generate differently (a renamed task, a new route) is reported as a `conflict`: yours is kept and the proposal is written under `.artifacts/verification/scaffold/<stamp>/` so you can compare the two. A newly found feature gets a new file beside the index; the audit then reports it `unlinked-map` until you link it. Exit code 1 means a problem that needs a person: no reusable check was found, a `verify` task is only inherited from a parent directory, a `verify` task runs `mise run verify` again, or mise is absent. Re-running on a completed repository reports every file `unchanged` and writes nothing. `--write` records copied skill files in `.agents/skills/.verification-provenance.json` with the source commit when `--revision` is a 40-character SHA (or this skill's git HEAD). ## 2b. Update from a named revision ```sh python3 <path-to-this-skill>/scripts/verify_scaffold.py --root . --update --revision <40-character-sha> ``` `--revision` must be an immutable commit in the skill git repository. `main`, `latest`, `HEAD`, and short SHAs are refused. Unchanged stock copies of `verify`, `evidence`, and `maintain-verification` may be replaced from that commit. A locally edited skill file is kept and listed as `conflict`. `VERIFY.md`, README facts, tasks, CI, and project-authored tests are never overwritten. Re-running the same revision after it is adopted writes nothing. Inspect and `--inspect` still write nothing and run no target command. The lifecycle is inspect, then a reviewed onboarding or update PR, then local and CI proof, then a human merge. New tasks see the adopted revision. In-flight tasks keep the `verification_policy` snapshot recorded at prepare. ## 3. Fill the map from observation Replace every `TODO(verify)` with what you observed: the purpose and each user entrypoint, sub-features and variants as extra scenario rows, how to drive it through the real UI/CLI/API, the observable states and side effects, the success / error / cancel / empty / persistence cases that apply (say in prose when one does not), the test or check that covers an `automated` row, and the gaps only a person can check. A row is `automated` only when its Driver names a test or command that exists and exercises it; a browser or desktop path with no installed driver stays `manual`. Keep the `Inventory: incomplete` label until a person has inventoried the whole product; a seed map never implies full coverage. Run the audit until it is clean: ```sh python3 .agents/skills/maintain-verification/scripts/verify_audit.py ``` ## 4. Prove it once 1. Launch in the repository's own isolation: its documented dev task, a temporary data directory, an ephemeral port, never the user's running instance or personal profile. 2. Readiness: `python3 .agents/skills/verify/scripts/verify_run.py --check`, then the readiness steps `VERIFY.md` lists (a health endpoint, a version line). 3. Run the contract: `python3 .agents/skills/verify/scripts/verify_run.py --base <merge-base>`; read the printed record path. 4. Drive one mapped feature end to end through its real entrypoint and capture evidence beside that record: ```sh python3 .agents/skills/verify/scripts/verify_capture.py --feature <id> http --expect-status 200 GET http://127.0.0.1:<port>/<route> python3 .agents/skills/verify/scripts/verify_capture.py --feature <id> run --expect-text 'Hello' -- <command> <args> ``` 5. Run the Teardown section, then confirm `.artifacts/verification/<run-id>/run.json` and its `evidence/` files still exist. A cleanup that eats the proof fails this step. 6. Fix what failed and run the teardown after every failed iteration so nothing is left running. If the base itself is broken, report the cause; do not paper over it in the contract. Report the run record path, the evidence paths, which feature you drove, and what stays `manual`. ## 5. Hand over the maintenance loop Point the reader at `.agents/skills/maintain-verification/SKILL.md`. Repository changes go through the project's ordinary PR and human-merge path; do not write into other clones and do not schedule an automatic routine. ## Attribution Adapted from the Cursor pstack `create-verification-skill` and Lauren Tan's `verify-atlas` example (MIT); see `references/ATTRIBUTION.md`.
From douglasjarquin/sum, MIT.
More from douglasjarquin/sum
- 0.567score
evidence
douglasjarquin
Capture truthful before/after evidence for one mapped scenario - real screenshots, short screencasts, CLI transcripts, or HTTP responses from the base and the…
SkillAFreshness: AClaude Code +9 - 0.565score
sum-factory
douglasjarquin
Start or run a factory from a markdown settings file.
SkillAFreshness: AClaude Code +9 - 0.565score
factory
douglasjarquin
Start or run a factory from a markdown settings file.
SkillAFreshness: AClaude Code +9 - 0.565score
verify
douglasjarquin
Run a distinct coordinator verification of a worker result.
SkillAFreshness: AClaude Code +9 - 0.565score
recap
douglasjarquin
Recap this session from saved history only.
SkillAFreshness: AClaude Code +9 - 0.565score
status
douglasjarquin
Reconcile /workspace/sum/inbox.md with live worker Bots.
SkillAFreshness: AClaude Code +9 - 0.565score
persist
douglasjarquin
Save briefs, questions, answers, and reports as files under /workspace/sum before chatting about them.
SkillAFreshness: AClaude Code +9 - 0.564score
dispatch
douglasjarquin
Hand approved work to one worker Bot.
SkillAFreshness: AClaude Code +9 - 0.561score
sweep
douglasjarquin
Put standing sweeps on a dedicated Sweep worker plus routine.
SkillAFreshness: AClaude Code +9 - 0.561score
sum-deliver
douglasjarquin
Independently inspect worker results, use existing verification tooling, and prepare a reviewable PR without autonomous merging.
SkillAFreshness: AClaude Code +9 - 0.561score
sum-develop
douglasjarquin
Improve sum itself from an isolated development checkout and lab state, without touching the installation that serves the live coordinator.
SkillAFreshness: AClaude Code +9 - 0.559score
sum-dispatch
douglasjarquin
Delegate explicitly approved work into an isolated Herdr task checkout with a durable worker brief.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 0
- Open issues
- 2
- Last commit (this folder)
- Sep 16, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/douglasjarquin/sum
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from douglasjarquin
- 0.567score
evidence
douglasjarquin
Capture truthful before/after evidence for one mapped scenario - real screenshots, short screencasts, CLI transcripts, or HTTP responses from the base and the…
SkillAFreshness: AClaude Code +9 - 0.565score
status
douglasjarquin
Reconcile /workspace/sum/inbox.md with live worker Bots.
SkillAFreshness: AClaude Code +9 - 0.565score
factory
douglasjarquin
Start or run a factory from a markdown settings file.
SkillAFreshness: AClaude Code +9 - 0.565score
persist
douglasjarquin
Save briefs, questions, answers, and reports as files under /workspace/sum before chatting about them.
SkillAFreshness: AClaude Code +9 - 0.565score
recap
douglasjarquin
Recap this session from saved history only.
SkillAFreshness: AClaude Code +9 - 0.565score
verify
douglasjarquin
Run a distinct coordinator verification of a worker result.
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 create-verification 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
create-verification skill. Agent Mart. https://agentmart.dev/skills/douglasjarquin-create-verification (last updated 2026-10-24)