Repository navigation
refactor: port pstack skills to harness-neutral tstack - #3
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ills Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5f0accf4b8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| ## Step 3, Spawn Reviewers | ||
|
|
||
| Launch all reviewers in a single message using the Task tool. Use the `interrogate reviewers` line in `~/.cursor/rules/pstack-models.mdc`, one reviewer per entry, extending or shrinking the Reviewer A/B/C labels below to the configured entry count. If the rule or that line is missing, use the table defaults. | ||
| Launch all reviewers in a single message, each a read-only host subagent on the strongest model. Same-family reviewers find different bugs only when they look from different angles, so each gets a lens: |
There was a problem hiding this comment.
Route interrogate reviewers through the bridge
When the host exposes only one model family, this launches every reviewer on the same strongest host model, so /interrogate no longer provides the multi-model review promised by its description. A repo-wide search confirms that skills/interrogate/references/bridge.md is not present, leaving no path to add cross-family reviewers; add the bridge and route panel members through it rather than collapsing the panel to one model.
AGENTS.md reference: AGENTS.md:L15-L15
Useful? React with 👍 / 👎.
| ``` | ||
|
|
||
| [`/setup-pstack`](../../skills/setup-pstack/SKILL.md) detects the models you have access to, asks for a reasoning budget, shows you each role (code delegates, judgment, the review panels), and asks what you want. Answer the questions. It writes `~/.cursor/rules/pstack-models.mdc`, a small rule every pstack skill reads. | ||
| `/setup-pstack` (not in tstack) detects the models you have access to, asks for a reasoning budget, shows you each role (code delegates, judgment, the review panels), and asks what you want. Answer the questions. It writes `~/.cursor/rules/pstack-models.mdc`, a small rule every pstack skill reads. |
There was a problem hiding this comment.
Replace the removed pstack setup flow
Any tstack user following this setup page is told to run /setup-pstack even though the line itself acknowledges that command is not shipped, and the surrounding steps install the Cursor-only pstack plugin rather than tstack. This makes the primary guide unusable outside Cursor and can install the wrong project inside Cursor; document npx skills add tacheshun/tstack and the host harness's model handling instead.
AGENTS.md reference: AGENTS.md:L3-L3
Useful? React with 👍 / 👎.
| "cursor-team-kit", "create-skill", "~/.cursor/", "poteto-mode", "poteto-agent", | ||
| "grok-", "gpt-5", "claude-opus-", | ||
| ]; | ||
| export const TERM_SCAN_ROOTS = ["skills"]; |
There was a problem hiding this comment.
Scan the guide for banned harness-specific terms
Because TERM_SCAN_ROOTS contains only skills, check.mjs reads docs/guide but never applies the banned-term check there. The current guide consequently passes while containing setup-pstack, pstack-models, cursor-team-kit, create-skill, and Cursor-only paths, so the required validator cannot prevent the user-facing portability regressions it is meant to catch; include the guide in the term-scan roots and cover it with a fixture.
AGENTS.md reference: AGENTS.md:L7-L8
Useful? React with 👍 / 👎.
Adds scripts/check.mjs (frontmatter, links, skill refs, banned Cursor-only terms and model slugs) and CI, renames poteto-mode to marius-mode, moves Comment Sicko into no-comments, and removes every Cursor-only tool name, path, model slug, and plugin dependency. Panels run on host subagents with distinct lenses; the cross-vendor bridge lands in the next PR. Stacked on #2.
Verified: node scripts/check.mjs → ok (45 skills); validator tests 10/10; marius-mode scripts 52/52 and strict typecheck; worktree-audit tested against a fake Claude Code transcript.
🤖 Generated with Claude Code