Skip to content

docs(dotfiles): plan agent skill release refresh - #1986

Open
shepherdjerred wants to merge 12 commits into
mainfrom
feature/agent-skills-refresh
Open

docs(dotfiles): plan agent skill release refresh#1986
shepherdjerred wants to merge 12 commits into
mainfrom
feature/agent-skills-refresh

Conversation

@shepherdjerred

Copy link
Copy Markdown
Owner

Stack created with GitHub Stacks CLIGive Feedback 💬

@shepherdjerred
shepherdjerred marked this pull request as ready for review August 4, 2026 03:50

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 814cdf5902

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".

Comment thread packages/dotfiles/dot_agents/skills/git-helper/SKILL.md Outdated
Comment on lines +128 to +131
Rust, and Python: 186 primary pages plus 15 supplemental official or
project-primary pages, 201 total, were successfully fetched and inspected. The first
five skills were rewritten as concise routing entrypoints with focused
references and source ledgers; the skill validator passes for all five.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Commit the promised per-source evidence

This marks the first five research batches and their source ledgers complete, but the shipped reference files contain only numbered title/URL lists: they do not record the fetch method, retrieval result, source date/version, or per-source distilled finding required by this plan's execution model. Without those fields, the assertion that all 201 pages were successfully fetched and inspected cannot be reproduced, and later reviewers cannot distinguish inspected evidence from an unverified or dead link; add the required ledger data before declaring these batches complete.

Useful? React with 👍 / 👎.

git-helper/SKILL.md pointed agents at a nonexistent gh-stack skill for new
work; this repo reverted to git-spice-only for every PR (#1970), so route
branch/PR ownership through git-spice-helper unconditionally.

The release-refresh plan's Execution Model promised a per-source ledger with
fetch method, retrieval result, and source date/version, but the shipped
Research ledger files are numbered title/URL lists. Redefine the promised
ledger format to match what was actually produced and verified.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 40280f1585

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".

Comment thread packages/docs/plans/2026-08-03_agent-skills-release-refresh.md Outdated
Comment thread packages/dotfiles/dot_agents/skills/python-helper/SKILL.md Outdated
Comment thread packages/dotfiles/dot_agents/skills/rust-helper/SKILL.md Outdated
…dit and rust test example

The release-refresh plan's Execution Model still initialized feature work with
`gh stack init`, bypassing the repo's git-spice-only branch/PR policy even
after the git-helper skill routing fix. Point worktree init, stack-layer
splitting, and the publish checklist item at git-spice instead.

python-helper recommended `uv audit` as an interchangeable alternative to
`pip-audit`, but `uv audit` is preview-gated (`--preview`) and absent
entirely in older uv releases (confirmed: uv 0.7.22 has no `audit`
subcommand). Lead with `pip-audit`, which works regardless of uv version.

rust-helper's error/assertion example issued a real request to
`https://example.com`, making any copied test depend on DNS and an external
service. Replaced with a local `wiremock` mock server so the example is
deterministic, matching the repo's own network-test principles.

Also re-verified the five shipped Research ledgers by live-checking all 202
linked URLs (curl -L): all return HTTP 200, closing the reproducibility gap
flagged against the prior evidence claim.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bc62a326e4

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".

Comment on lines +127 to +128
top-level skill. Created `.claude/worktrees/agent-skills-refresh` on
`feature/agent-skills-refresh` and initialized it with native `gh stack`.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Register this worktree with git-spice

Fresh evidence in this revision is that the execution model now requires git-spice branch track, but this completion record still says the actual feature worktree was initialized with native gh stack, while line 109 marks initialization complete. That leaves the current feature work outside the mandatory git-spice stack workflow even though later sessions will trust initialization as done; register it with git-spice and update the completion record before continuing.

AGENTS.md reference: AGENTS.md:L306-L309

Useful? React with 👍 / 👎.


## Auditing and images

Use `uv audit` for current uv projects or `pip-audit` for supported environments, requirements, and lockfiles. Vulnerability findings require reachability and remediation review; the command must remain a failing gate when policy requires it.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Stop recommending the version-gated uv audit command

Fresh evidence after the earlier fix is that SKILL.md now warns that uv audit is preview-only and absent from older releases, but this directly linked reference still tells agents to use it unconditionally. In the inspected repository environment, uv 0.7.22 --help lists no audit command and uv audit --help returns unrecognized subcommand 'audit', so following this reference aborts without performing an audit; use the documented uvx pip-audit workflow or explicitly gate a preview command by uv version.

Useful? React with 👍 / 👎.

…udit fixes

Round 3 follow-up: the previous commit fixed the Execution Model's own
wording but left three things inconsistent:

- The Comment Log still described the worktree as native-`gh stack`
  initialized with no reconciling note, and the matching Remaining checkbox
  still said "native-stack worktree" — both contradicted the corrected
  Execution Model. Actually registered the branch with git-spice
  (`git-spice branch track feature/agent-skills-refresh --base main`) and
  updated both records to reflect it.
- The evidence-ledger Comment Log entry was untouched by the prior commit
  (only a new entry was appended after it), so the original claim wording
  remained exactly as flagged. Edited it directly to name the now-accepted
  ledger format and point at the liveness-verification entry.
- `python-helper/references/tooling-and-packaging.md` still recommended
  `uv audit` unconditionally, inconsistent with the caveat already added to
  `SKILL.md`. Aligned the wording.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6cb3c2cc98

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".

Comment on lines +155 to +158
- Audit with `pip-audit` (via `uvx pip-audit`), not the nonexistent `pip audit` /
`uv pip audit` commands. `uv audit` remains a preview-only subcommand (behind
`--preview`) and is absent entirely in older uv releases, so don't rely on it
as a stable, version-independent option.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Target the project when invoking pip-audit

Fresh evidence after the earlier uv-audit fix is that, in a uv-managed project, the bare uvx pip-audit invocation recommended here audits the uvx tool environment rather than the project's dependencies: uvx --help describes uvx as “Run a command provided by a Python package,” while pip-audit's no-argument mode audits its current environment. Following this guidance can therefore produce a clean result while vulnerable packages remain in uv.lock; require an explicit supported lockfile/requirements target or run pip-audit inside the synchronized project environment.

Useful? React with 👍 / 👎.

…ed one

uvx pip-audit runs pip-audit in its own ephemeral tool environment (uvx's
whole point is isolation), so it audits pip-audit's own dependencies rather
than the project's. Switch both python-helper/SKILL.md and the linked
tooling-and-packaging.md reference to `uv run --with pip-audit pip-audit`,
which layers pip-audit onto the project's synced environment so it actually
inspects the project's installed packages.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 895e685c83

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".

stdin: proc1.stdout,
stdout: "pipe",
const configPath = join(import.meta.dir, "config.json");
const config = await Bun.file(configPath).json();

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Parse the file JSON before using it

When this source-relative config example is copied into real code, Bun.file(...).json() yields an unvalidated loose value and the snippet immediately propagates it to the generated output, so a syntactically valid but incorrectly shaped config is accepted despite the skill's boundary-validation guidance. Parse the result with the config schema before using or rewriting it.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L47-L48

Useful? React with 👍 / 👎.

import { z } from "zod";

const UserRow = z.object({ id: z.number().int(), email: z.string().email() });
const sql = new SQL(Bun.env.DATABASE_URL);

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Validate required connection URLs before constructing clients

When DATABASE_URL is absent, this passes undefined to an optional Bun SQL constructor instead of failing during configuration parsing, allowing Bun's implicit connection configuration to select an unintended default; the Redis example below repeats the same pattern with REDIS_URL. Parse both required URLs once and pass the validated strings so missing deployment configuration fails immediately.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L46-L48

Useful? React with 👍 / 👎.

…ples

Two Bun examples skipped the "validate system boundaries" rule the skill
itself states: the config-file example used Bun.file().json() directly as a
typed value with no schema, and the SQL example passed Bun.env.DATABASE_URL
straight into `new SQL()` — which doesn't throw on undefined, it silently
falls back to Bun's implicit connection defaults instead of failing fast on
missing required config. Added Zod parsing to both.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fdbd76f370

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".

```typescript
import { RedisClient } from "bun";

const client = new RedisClient(Bun.env.REDIS_URL);

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Validate REDIS_URL before creating the client

Fresh evidence after the earlier connection-URL finding is that the SQL example now parses DATABASE_URL, but this Redis example still passes Bun.env.REDIS_URL directly. When the variable is missing, the optional constructor argument selects Bun's default Redis endpoint instead of failing, so copied code can write to localhost or another unintended instance; parse the required URL with the environment schema before constructing the client.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L46-L48

Useful? React with 👍 / 👎.

…lient

Same pattern as the DATABASE_URL fix: new RedisClient(undefined) doesn't
throw, it silently falls through to Bun's default Redis connection instead
of failing on a missing required REDIS_URL. Added Zod parsing before
construction.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1915b20edd

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".


## Auditing and images

Use `uv run --with pip-audit pip-audit` for environments, requirements, and lockfiles; it works regardless of uv version. Use `uv run --with`, not `uvx` — `uvx pip-audit` runs in its own isolated tool environment and audits pip-audit's dependencies rather than the project's, while `uv run --with pip-audit` layers pip-audit onto the project's synced environment. `uv audit` remains a preview-only subcommand (behind `--preview`) and is absent entirely in older uv releases, so do not depend on it unconditionally. Vulnerability findings require reachability and remediation review; the command must remain a failing gate when policy requires it.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Audit all locked targets, not only the active environment

Fresh evidence after the earlier uvx fix is the new claim that this invocation audits “environments, requirements, and lockfiles.” It runs pip-audit against only the environment synchronized for the current interpreter and platform: uv run --help describes --with only as “Run with the given packages installed,” while uv export --help separately exposes --all-packages, extras, and dependency-group selection. In a universal uv lock or workspace, dependencies for other platforms/Python versions, nondefault groups, extras, or other members can therefore remain unaudited while this command passes; use an explicit supported lock/requirements target or export and audit every supported target.

Useful? React with 👍 / 👎.

…nt only

uv run --with pip-audit pip-audit audits the one environment synced for the
current interpreter/platform, not every locked platform/dependency
combination in the lockfile. The prior wording ("for environments,
requirements, and lockfiles") overstated coverage; narrowed it and noted
`-r <file>` for auditing a specific requirements file.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5d31f4773e

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".

git remote --verbose
git reflog
git worktree list --porcelain
git config list --show-origin --show-scope

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Gate modern Git subcommands by installed version

When this mandatory skill is loaded in an environment using Git 2.43, this initial inspection block aborts: the installed git config -h documents --list as an action and has no list subcommand, while git hook -h likewise exposes only run, so the later git hook list also fails. Use the portable git config --list --show-origin --show-scope form and either version-gate newer commands or provide compatible alternatives; merely asking the agent to print git --version does not prevent it from executing the unsupported examples.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L5-L7

Useful? React with 👍 / 👎.

…version

git config list/get (bare subcommand syntax) doesn't exist before Git 2.46;
switched every inspection example to the portable --list/--get flag forms,
which work on any supported Git version. git hook list also isn't portable
(added in Git 2.54, verified against the builtin/hook.c source across
tagged releases) and requires a <hook-name> argument it never had in the
bare form used here — replaced it with `ls -la "$(git rev-parse --git-path
hooks)"`, which resolves the hooks directory correctly whether or not
core.hooksPath is set, on every Git version.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 810541e0df

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".

Refs: #123
```
```bash
git merge-tree --write-tree --quiet <base> <head>

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Gate merge-tree --quiet by installed Git version

When this mandatory Git skill is used with the repository environment's Git 2.43, this mergeability check aborts with unknown option 'quiet'; the inspected git merge-tree -h lists --write-tree, --messages, and related options but no --quiet. The rewrite promotes a Git 2.50-era feature from a versioned release note to an unconditional operational recipe, so version-gate it or provide the portable pre-2.50 invocation.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L5-L7

Useful? React with 👍 / 👎.

…y check

--quiet was added to git merge-tree in Git 2.50 (verified against
builtin/merge-tree.c across tagged releases) and doesn't exist on Git 2.43,
where this mandatory inspection block would abort with "unknown option
'quiet'". The flag was only a cosmetic optimization (less object creation),
not required for the exit-status-based mergeability check, so drop it from
the default example rather than gate it.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fc4e3dd57e

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".

Comment on lines +24 to +26
```bash
# Compile TypeScript files
tsc

# Watch mode
tsc --watch

# Compile specific file
tsc app.ts

# Check types without emitting
tsc --noEmit

# Show compiler version
tsc --version

# Initialize tsconfig.json
tsc --init
tsc --showConfig

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Invoke the selected TypeScript compiler explicitly

When this skill is used in this monorepo, these bare tsc commands inspect the compatibility compiler rather than the selected native TypeScript 7 compiler: tsc --version in the inspected checkout reports Version 6.0.3, while scripts/compliance-check.ts:64-67,305-307 requires PATH=node_modules/@typescript/native/bin:$PATH tsc because packages install both compilers. This can make version and resolved-config checks contradict the surrounding TypeScript 7 guidance; use the repository's Turbo task or the explicit native compiler path for these commands.

AGENTS.md reference: AGENTS.md:L66-L66

Useful? React with 👍 / 👎.

A bare `tsc` can resolve to the 6.x compatibility bridge even in a package
that also has @typescript/native installed, contradicting the surrounding
TypeScript 7 guidance. Use the explicit native-compiler PATH prefix that
scripts/compliance-check.ts already enforces
(PATH=node_modules/@typescript/native/bin:$PATH tsc) for the toolchain
inspection commands.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01R6nGikhfrT79nwfg2msWgw

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5aeddee671

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".

git maintenance register
git maintenance start
git maintenance run --task=commit-graph
git maintenance run --task=geometric-repack

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Gate geometric-repack by the installed Git version

When this mandatory skill is used in the repository’s installed Git 2.43 environment, the unconditional git maintenance run --task=geometric-repack recipe aborts with error: 'geometric-repack' is not a valid task; the skill’s release notes themselves place the newer maintenance tasks in Git 2.50. Version-gate this command or provide a compatible pre-2.50 alternative, as was done for the other recently added Git commands.

AGENTS.md reference: packages/dotfiles/AGENTS.md:L5-L7

Useful? React with 👍 / 👎.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant