Skip to content
Skills Library
AI & agents · Testing

verify-pstack

/verify-pstack

Use when verifying this pstack Grok plugin, leftover Cursor harness call sites may still be live (leftover-scanner is Cursor harness token doctor), plugin doctor is in doubt, refresh-hygiene, upstream pin, upstream recipe, or release-tag checks are needed, or an agent reaches for grok plugin validate, grok inspect, or pytest as a substitute for scripts/verify_harness.py.

Repository
pstack ↗
Licence
MIT
Last repository change
26.10.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
10
Scripts
2
HTML comments
0
URLs in SKILL.md
1
Deprecation

No signals detected. Repository changed on 26.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 2 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: scripts/absorbed.py, scripts/verify.py

What it does and when to use it

Use when verifying this pstack Grok plugin, leftover Cursor harness call sites may still be live (leftover-scanner is Cursor harness token doctor), plugin doctor is in doubt, refresh-hygiene, upstream pin, upstream recipe, or release-tag checks are needed, or an agent reaches for grok plugin validate, grok inspect, or pytest as a substitute for scripts/verify_harness.py.

Original author description

Use when verifying this pstack Grok plugin, leftover Cursor harness call sites may still be live (leftover-scanner is Cursor harness token doctor), plugin doctor is in doubt, refresh-hygiene, upstream pin, upstream recipe, or release-tag checks are needed, or an agent reaches for grok plugin validate, grok inspect, or pytest as a substitute for scripts/verify_harness.py.

How it works

  1. Launch
  2. Doctor
  3. Drive
  4. Proof bar
  5. Evidence
  6. Cleanup
  7. Helpers
  8. Refused paths
  9. Red flags
Original SKILL.md headings · 147 lines, 1,267 words
SKILL.md preview
---
name: verify-pstack
description: "Use when verifying this pstack Grok plugin, leftover Cursor harness call sites may still be live (leftover-scanner is Cursor harness token doctor), plugin doctor is in doubt, refresh-hygiene, upstream pin, upstream recipe, or release-tag checks are needed, or an agent reaches for grok plugin validate, grok inspect, or pytest as a substitute for scripts/verify_harness.py."
disable-model-invocation: true
---

# Verify pstack

Short-lived CLI plugin. No server. Drive it with `.grok/skills/verify-pstack/scripts/verify.py`. Do not improvise doctor.

Plugin doctor lives at `.grok/skills/verify-pstack/`. Do not ship it under `skills/`. `plugin.json` lists `./skills/` and must not list `.grok/skills`. Do not write `.claude/skills`. Do not write `~/.grok/skills`.

The map lives at `features/`, next to this file. Do not read OpenSpec for drive recipes. Do not look under `references/features/`.

## Launch

Work in a pstack plugin checkout. `plugin.json` name is `pstack`. `scripts/verify_harness.py` exists.

Need `python3` and `grok` on `PATH`. Refresh-hygiene also needs `uv`. Do not run `mise use -g` to get them.

Each drive is its own process. Isolation is `--run-id`, not `--checkout`. There is no CDP checkout flag. Two runs may share a checkout when they use different `--run-id` values. There is no long-lived instance to keep alive. Refuse a second drive that would share live `~/.grok/skills` or the user's overlay dest.

Ready means leftover scanner printed `PASS` and exited 0. That PASS is checkout token cleanliness, not catalog identity. Run doctor until leftover PASS is true.

## Doctor

Leftover scanner is required doctor. Run it. No substitutes. leftover-scanner is not git leftover clones; isolation reclaim is the worktree-cleanup playbook.

```bash
python3 .grok/skills/verify-pstack/scripts/verify.py doctor --root .
```

That runs `python3 scripts/verify_harness.py`, then `grok plugin validate .`. Leftover stdout must start with `PASS` and include `playbooks: 22 named + opening-a-pr`, `principles: 24`, and `plugin.json name: pstack`. Leftover exit 0. Validate stdout contains `Plugin manifest is valid.` Exit 0.

`python3 scripts/check_plugin_agents.py` is not leftover scanner. It is the enable check for workflow `agent_type` `pstack:swarm-workers`. It reads inspect `plugins[].path` and `skills[]` `collidesWith`. `[compat.claude] skills=false` does not hide `~/.grok/skills`. If inspect lacks `pstack:swarm-workers`, Grok is not loading https://github.com/tommy-ca/pstack. `grok plugin marketplace add` is catalog-only. Do not treat it as plugin install.

Later `drive` on the same `--run-id` refuses if leftover scanner did not PASS. `drive leftover-scanner` may run the scanner without a prior doctor log. Other features still require that leftover `PASS` in this run's doctor evidence.

**No exceptions.**

- Do not treat `grok plugin validate` as leftover scanner.
- Do not treat `grok inspect --json` as leftover scanner. `inspect` `plugins[].enabled` is trust.
- Do not treat `pytest` or `tests/test_verify_harness.py` as leftover scanner. Those are not the tree walk.
- Do not skip leftover scanner because TEST-PLAN.md says it is not an EDITH pass gate. TEST-PLAN is the live-CLI gate plan. This skill still requires the leftover scanner as doctor.
- Do not run validate first and call doctor done if leftover scanner did not print `PASS`.

Doctor also fails if `<root>/.grok/workflows` exists. Do not create that directory.

Re-run doctor after any failed drive, and before the first drive of a run.

| Excuse | Reality |
|---|---|
| `grok plugin validate` plus inspect is doctor | Leftover scanner is required doctor. Validate is the companion only. Inspect is not leftover scanner. `scripts/check_plugin_agents.py` is the enable check. |
| pytest already covers leftover tokens | `tests/test_verify_harness.py` is not `scripts/verify_harness.py` walking the tree. Run the scanner. |
| TEST-PLAN says verify-harness is not a pass gate | EDITH live-CLI gates are a different check. This doctor still requires leftover `PASS`. |
| Static scan is not a user path | The operator path is `python3 scripts/verify_harness.py`. Drive it. |
| leftover-clone or worktree-drop leftover_count | leftover-scanner is Cursor harness token doctor. Isolation reclaim is worktree-cleanup. |
| I will run leftover scanner later | Doctor first. A drive without leftover `PASS` is invalid. |

## Drive

```bash
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature leftover-scanner
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature upstream-pin
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature absorbed-intent
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature upstream-recipe
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature refresh-hygiene
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature release-tag
```

Omit `--feature` to drive every mapped feature. That walk follows `features/README.md` top to bottom. It is the Full sweep. Driving one convenient feature is not a sweep. That path requires leftover `PASS` in this run's doctor evidence first. `drive leftover-scanner` may run without that log. Other features on the same `--run-id` refuse until leftover scanner PASSed.

Read `features/README.md`, then the feature file. Drive the listed commands through `verify.py`. Do not paste `refresh_hygiene.py --apply-skills`. Do not run `scripts/release.sh` to completion from nested grok.

## Proof bar

Do not submit "look, it opens" captures. `grok plugin validate` plus inspect is not leftover scanner. A leftover `PASS` line, a 40-hex pin, `PASS absorbed-intent`, `--recipe` stdout needles plus argv without `--log`, a hygiene TSV, or `PASS tests/test_release.py` counts only when it is the mapped operator path with stdout, stderr, and exit code in the same evidence directory.

- Drive the operator command through `verify.py`. Do not call internal Python functions as the proof.
- Run doctor first. A capture without leftover `PASS` in this run's doctor log is not evidence.
- Read the feature file. Exercise every reachable entry point it lists, and the success, cancel, error, empty, and persistence paths the change can affect.
- For a broad regression, walk `features/README.md` top to bottom. That is leftover-scanner, upstream-pin, absorbed-intent, upstream-recipe, refresh-hygiene, then release-tag.
- Show the trigger command and the stable end state in the same evidence directory.
- Verify side effects, not only stdout. Hygiene must leave `~/.grok/skills/reflect` unchanged and write `--host-script` dest under `/tmp/`.

## Evidence

Directory: `/tmp/verify-pstack-evidence-<runid>/`. `verify.py` prints `run_id` and `evidence` on PASS.

Keep command, stdout, stderr, and exit code for every step. Leftover proof is the `PASS` line plus playbook and principle counts. Pin proof is the 40-hex SHA that matches `UPSTREAM`. Recipe proof is `--recipe` stdout needles plus argv without `--log`. Hygiene proof is the TSV header, the `cp -- ` line, and an unchanged live skills fingerprint. Release proof is `PASS tests/test_release.py`.

Cleanup must not delete this directory.

## Cleanup

```bash
python3 .grok/skills/verify-pstack/scripts/verify.py cleanup --run-id <runid>
```

Removes `/tmp/verify-pstack-scratch-<runid>/` only. Confirm the evidence directory still exists after cleanup. Do not kill by process name. Do not `git worktree remove`. Isolation reclaim is the worktree-cleanup playbook, not verify cleanup. Do not delete leftover-worktree rows for OPEN PRs. Do not `rm` `/tmp/verify-pstack-evidence-<runid>/`.

## Helpers

`.grok/skills/verify-pstack/scripts/verify.py` is executable. This skill ships that driver. The verification-skill-example omits its driver on purpose. Do not omit this helper.

From a checkout that already has the skill:

```bash
python3 .grok/skills/verify-pstack/scripts/verify.py doctor --root .
python3 scripts/check_plugin_agents.py
python3 .grok/skills/verify-pstack/scripts/verify.py drive --root . --feature leftover-scanner
python3 .grok/skills/verify-pstack/scripts/verify.py run --root .
python3 .grok/skills/verify-pstack/scripts/verify.py cleanup --run-id <runid>
```

`run` is doctor, then every feature (or `--feature`), then cleanup. Evidence survives `run`.

Pass `--root` when the script is not inside `.grok/skills/verify-pstack/scripts/`. Pass `--run-id` to isolate concurrent runs. Do not pass `--checkout`.

## Refused paths

`verify.py` refuses these. Do not run them by hand during a verify run either.

- `--log` (sync-only fetch; not a verify path)
- `--apply-skills` (default dest is `~/.grok/skills`; live `reflect` may be a symlink)
- `scripts/release.sh` to completion when `__GROK_INSIDE_BWRAP` is set
- `mise use -g`
- `git worktree remove` of OPEN PR worktrees. Isolation reclaim is worktree-cleanup, not verify.
- creating plugin `.grok/workflows`
- `chmod` on overlay dests
- writes under `~/.grok/skills`

Hygiene `--host-script` and dry-run must pass `--skills` under `/tmp/verify-pstack-scratch-<runid>/`. Default `--skills` is live dest. Do not use the default.

## Red flags

Stop. Run `verify.py doctor`. Start over.

- Doctor log has validate, inspect, or pytest and no leftover-scanner `PASS`
- Evidence missing `doctor/leftover-scanner/stdout.txt`
- `--log` on `verify.py`
- `--apply-skills` without a tmp `--skills`
- dest path contains `/.grok/skills/`
- `scripts/release.sh` ran inside nested grok
- `mise use -g`
- `git worktree remove`
- `chmod` on an overlay dest
- cleanup deleted `/tmp/verify-pstack-evidence-*`
- a proof that only shows the process started

Keep the map honest with `/maintain-verification-skill`.
Files in this package (10)
  • SKILL.md 10 KB
  • features/README.md 4 KB
  • features/absorbed-intent.md 3 KB
  • features/leftover-scanner.md 4 KB
  • features/refresh-hygiene.md 3 KB
  • features/release-tag.md 3 KB
  • features/upstream-pin.md 2 KB
  • features/upstream-recipe.md 2 KB
  • scripts/absorbed.py 5 KB
  • scripts/verify.py 22 KB

Versions

The database has no other skills named verify-pstack.

Database snapshot: 13.11.2026 · package version 8f2c65d331e8 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100setup-pstack-grok
    Grok Build adapter for pstack setup. Detects the models the task tool accepts and the live grok-build effort enum, then writes ~/.grok/pstack-models.toml plus ~/.grok/roles/pstack:<role>.toml. Dispatched from /setup-pstack when profiles/grok.json#setup names this skill; use it…
  • 95setup-pstack
    Detect the active agent harness and configure which models pstack uses per role. Writes ~/.pstack/models.md, which the skills read as an override on the harness defaults. Use for /setup-pstack, "configure pstack models", or changing pstack's model choices.
  • 100orca-pstack
    Run a pstack role on the harness that has its model. Use when poteto-mode, arena, swarm, how, why, architect, interrogate, reflect, show-me-your-work, or setup-pstack is about to spawn a subagent whose model this harness cannot run. Dispatches an Orca orchestration worker on any…
  • 95verify-open-pstack
    Verify Open Pstack through installed Codex and Claude Code skills and its shipped Bun CLIs. Use for real behavior proof, routing or setup changes, orchestration bookkeeping, and release evidence.
  • 95setup-pstack
    Configure which host-supported roles or models pstack uses per workflow role. Detects the live host inventory and writes portable project-local configuration. Use for /setup-pstack, "configure pstack models", or changing pstack's role choices.
  • 100setup-pstack
    Configure which models pstack uses per role and at what reasoning budget. Offers every agent Orca orchestration can launch, not only the CLIs on this machine, then writes one agent, model, and effort per role. Use for /setup-pstack, "configure pstack models", "pstack budget", or…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.