Skip to content

chore(skills): add user forum What's New recap skill - #6314

Merged
rh-hemartin merged 14 commits into
mainfrom
chore/skills-user-forum-whats-new
Sep 10, 2026
Merged

rh-hemartin merged 14 commits into
mainfrom
chore/skills-user-forum-whats-new

Conversation

@ascerra

@ascerra ascerra commented Aug 18, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Adds a contributor skill for the weekly Fullsend user-forum What's New recap: gather shipped work, rank it, and produce a short talk track with clickable examples. Audience already has GitHub release notes; this is the 4–6 minute verbal pass, not a changelog dump.

Related Issue

None — intentionally untracked forum-host tooling. Do not auto-close unrelated issues on merge.

Changes

  • skills/user-forum-whats-new/SKILL.md — workflow, ranking bands, newness / experience / talkable filters, Released vs On main output, HTML-escaping requirements, and anti-patterns
  • skills/user-forum-whats-new/scripts/gather.py — timezone-aware release/PR gathering, per-repo release cutoffs, strict date/window validation, truncation reporting, fail-closed JSON parsing, and bounded gh calls
  • skills/user-forum-whats-new/scripts/gather.sh — repo-root wrapper
  • skills/user-forum-whats-new/scripts/gather_test.py — offline unit tests for windows, classification, fetch argv/timeouts, parsing, truncation, and error paths
  • Makefile — runs the new gather tests through script-test

This is intentionally org forum-host tooling, not a sandbox agent skill, so it is not cross-linked from BYOA/customizing-with-skills guides. The skill runs from a fullsend checkout; .cursor/skills and .claude/skills symlinks make it discoverable.

Testing

  • pre-commit run on staged files (ruff, ty, bandit, shellcheck, markdown links, secrets)
  • python3 skills/user-forum-whats-new/scripts/gather_test.py (34 tests)
  • git diff --check

Checklist

  • PR title follows Conventional Commits
  • Commits are signed off (DCO)
  • Tests added/updated for new or modified logic
  • I wrote this contribution myself and can explain all changes in it

@ascerra
ascerra requested a review from a team August 18, 2026 11:09
@fullsend-ai-review

fullsend-ai-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:10 AM UTC · Completed 11:27 AM UTC

Commit: e7ef1d9 · View workflow run →

@qodo-code-review

Copy link
Copy Markdown

PR Summary by Qodo

Add user forum What's New recap skill

✨ Enhancement 📝 Documentation 🕐 20-40 Minutes

Grey Divider

AI Description

• Adds a ranked, example-first workflow for weekly Fullsend user-forum recaps.
• Gathers releases and merged pull requests across Fullsend repositories by date window.
• Produces concise Google Docs-ready talk tracks instead of duplicating release notes.
Diagram

graph TD
  Host["Forum Host"] -->|requests| Skill["Recap Skill"] -->|runs| Gather["Gather Script"]
  Gather -->|queries| Fullsend["Fullsend Repo"] -->|returns| JSON["Candidate JSON"] -->|feeds| Rank["Scoring Filters"] -->|produces| HTML["Docs HTML"]
  Gather -->|queries| Agents["Agents Repo"] -->|returns| JSON
Loading
High-Level Assessment

The repository-local skill paired with a small gh-based gathering script is appropriate for host-facing contributor tooling: it remains discoverable through existing skill symlinks, keeps editorial judgment in readable guidance, and automates only repeatable collection work. A hosted service or GraphQL application would add maintenance and deployment overhead without improving this weekly, human-reviewed workflow.

Files changed (2) +389 / -0

Enhancement (1) +126 / -0
gather.shGather weekly releases and merged pull requests +126/-0

Gather weekly releases and merged pull requests

• Adds a strict Bash utility that validates date arguments and dependency availability, then queries GitHub for releases and merged pull requests in 'fullsend-ai/fullsend' and 'fullsend-ai/agents'. Embedded Python filters releases to the inclusive date window and emits normalized candidate JSON.

skills/user-forum-whats-new/scripts/gather.sh

Documentation (1) +263 / -0
SKILL.mdDefine the weekly user-forum recap workflow +263/-0

Define the weekly user-forum recap workflow

• Introduces a contributor skill for collecting, filtering, scoring, and presenting weekly shipped changes. It establishes example requirements, audience and talkability filters, ranking bands, Google Docs HTML output rules, and anti-patterns that prevent changelog-style recaps.

skills/user-forum-whats-new/SKILL.md

@qodo-code-review

qodo-code-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📜 Skill insights (0)

Grey Divider


Action required

1. Forum boundary includes stale work ✓ Resolved 🐞 Bug ≡ Correctness
Description
in_window() discards release timestamps and compares calendar dates, while PR search also receives
date-only bounds, so every Tuesday run includes changes made before the previous Tuesday's 08:00
forum. Those stale changes can be ranked and presented as newly shipped even though the skill
explicitly defines the window as starting at the prior meeting.
Code

skills/user-forum-whats-new/scripts/gather.sh[R81-82]

+    day = iso[:10]
+    return since <= day <= until
Relevance

●● Moderate

Timestamp-boundary correctness is plausible, but history provides no close weekly-window precedent.

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The skill defines the start as the previous meeting's 08:00 America/New_York timestamp and requires
same-day releases only when published after that meeting. The implementation instead takes
iso[:10] and performs an inclusive date comparison; the PR queries likewise use only the supplied
dates.

skills/user-forum-whats-new/SKILL.md[57-63]
skills/user-forum-whats-new/scripts/gather.sh[55-63]
skills/user-forum-whats-new/scripts/gather.sh[78-82]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The gatherer implements the forum window as inclusive calendar dates, causing changes before the previous Tuesday meeting to be included. Use timezone-aware timestamps matching the documented America/New_York meeting boundary for both release filtering and PR search.

## Issue Context
The skill defines `SINCE` as the previous meeting's start, but the script accepts only dates and truncates release timestamps to their date component. Preserve the simple CLI if desired, but derive and compare exact UTC boundaries with correct daylight-saving handling.

## Fix Focus Areas
- skills/user-forum-whats-new/SKILL.md[57-70]
- skills/user-forum-whats-new/scripts/gather.sh[33-37]
- skills/user-forum-whats-new/scripts/gather.sh[55-63]
- skills/user-forum-whats-new/scripts/gather.sh[78-82]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools



Remediation recommended

2. Release bodies lose features ✓ Resolved 🐞 Bug ≡ Correctness
Description
The gatherer emits only the first 8,000 characters of each release body, so features listed later
disappear from candidate discovery. This is unrecoverable when their PRs merged before SINCE, even
though the skill explicitly says those release-body features must still be scored.
Code

skills/user-forum-whats-new/scripts/gather.sh[R101-102]

+                # Candidate list only — never paste this as the recap.
+                "body": body[:8000],
Relevance

●●● Strong

Silent truncation loses required candidate data; accepted history consistently favors preventing
incomplete gathered results.

PR-#6138
PR-#3610

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The workflow states that output includes the changelog body and that release-body features remain
candidates when their PRs predate SINCE. The emitted body[:8000] silently discards everything
after that boundary.

skills/user-forum-whats-new/SKILL.md[75-80]
skills/user-forum-whats-new/scripts/gather.sh[92-103]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The JSON truncates release bodies at 8,000 characters, potentially deleting the only record of in-scope features whose PRs merged before the search window. Emit complete bodies or provide a lossless follow-up mechanism.

## Issue Context
The skill specifically treats release changelogs as candidate sources and requires scoring features from them even when no corresponding PR appears in `merged_prs`. Any output-size safeguard must not silently remove those candidates.

## Fix Focus Areas
- skills/user-forum-whats-new/SKILL.md[75-80]
- skills/user-forum-whats-new/scripts/gather.sh[92-103]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


3. Release pagination drops candidates ✓ Resolved 🐞 Bug ≡ Correctness
Description
Each releases endpoint fetches only the newest 20 releases and never follows pagination, so an
in-window release outside that first page is silently missing. This especially breaks retrospective
or longer-window runs because filtering occurs only after the limited response is downloaded.
Code

skills/user-forum-whats-new/scripts/gather.sh[R52-53]

+gh api "repos/fullsend-ai/fullsend/releases?per_page=20" >"$TMP/rel-fullsend.json"
+gh api "repos/fullsend-ai/agents/releases?per_page=20" >"$TMP/rel-agents.json"
Relevance

●●● Strong

Accepted history favors preventing silent API omissions; pagination-cap rejection is a different
contract context.

PR-#6138
PR-#3610

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The script makes one request per repository with per_page=20, then filters only those returned
arrays. The workflow promises releases from the requested window and contains no documented
20-release constraint.

skills/user-forum-whats-new/scripts/gather.sh[52-53]
skills/user-forum-whats-new/scripts/gather.sh[84-93]
skills/user-forum-whats-new/SKILL.md[75-80]
PR-#2617

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Release collection is limited to one 20-item API page, allowing valid in-window releases to disappear silently. Fetch all required pages and combine them into the arrays consumed by Python.

## Issue Context
The date filtering happens locally after retrieval, so the script cannot know that an omitted API page contains an in-window release. Ensure paginated `gh api` output remains valid JSON for the existing loader, for example by slurping and flattening pages.

## Fix Focus Areas
- skills/user-forum-whats-new/scripts/gather.sh[52-53]
- skills/user-forum-whats-new/scripts/gather.sh[84-93]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


Grey Divider

Context sources
✅ Compliance rules (platform): 56 rules

Grey Divider

Tip of the day
💡 Did you know, you can turn these tips off under Display preferences

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Grey Divider

Qodo Logo

@codecov

codecov Bot commented Aug 18, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@fullsend-ai-review

fullsend-ai-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes before merge.

Low

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md:8 — The prior WebFetch grant flagged in the previous review has been removed, and the remaining allowed-tools grants (gather scripts, read-only gh subcommands, path-prefixed xdg-open/open) match the documented workflow. One gap remains: Write is still unscoped — SKILL.md documents it exists only to write /tmp/fullsend-whats-new-YYYY-MM-DD.html, but the grant permits writing any path. No linked issue/ADR authorizes the broader grant.
    Remediation: Prefer a small helper invoked via Bash that only writes under /tmp/fullsend-whats-new-*.html (path allowlist), or emit HTML on stdout like peer gather skills, instead of granting unscoped Write.

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py:119 — No test covers draft release filtering. classify() still skips releases with draft: True (gather.py:186) before in-window/cutoff logic, but TestClassifyPerRepo only exercises draft: False (and prerelease) cases. A regression that removes the draft-skip guard could let a draft with published_at set the per-repo cutoff or appear in releases[].
    Remediation: Add a classify() test with an in-window draft: True release (with a published_at that would otherwise win max(times)) and assert it is omitted from releases[] and does not set release_cutoff_utc.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes.

Medium

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md:6 — The allowed-tools frontmatter grants capabilities beyond the minimum required: (1) Write is unscoped — the skill documents it is granted solely for writing /tmp/fullsend-whats-new-YYYY-MM-DD.html (line 71), yet the permission permits writing to any path. (2) WebFetch is granted with no documented use case in any of the 7 workflow steps. Note: the direct gh Bash permissions (gh search, gh pr view, etc.) are justified by step 3's ad-hoc example-hunting workflow and are correctly scoped.
    Remediation: Remove WebFetch unless a use case is documented. Consider documenting the Write scope limitation (allowed-tools does not support path-scoped Write) or using stdout output matching peer skills.

Low

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py — No test covers draft release filtering. The production code in classify() skips releases with draft: True, but no test verifies this behavior. A regression removing the draft-skip guard could cause draft releases to set the cutoff timestamp or appear in the output.

  • [scope-vs-convention] skills/user-forum-whats-new/SKILL.md — AGENTS.md line 22 conditionally requires checking user-facing guides for cross-reference opportunities when adding a new skill. The PR body explicitly states this is "intentionally not cross-linked" because it's a forum-host skill. The instruction is conditional ("if the new skill fills a gap in those guides"), and peer contributor skills (topissues, nextwork) are not cross-referenced in those guides either.

  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py — Uses sys.stderr.write() for error/warning output (6 call sites outside the two early guards) and raise SystemExit(main()) as the entry point, while both peer skill scripts (topissues.py, nextwork.py) uniformly use print(..., file=sys.stderr) and sys.exit() / bare main() respectively. The two early guard clauses have a legitimate reason for sys.stderr.write/sys.exit since they run before 3.11-dependent imports. Functionally equivalent — style preference only.

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — The shell wrapper is a novel sub-pattern for Python-based skills in skills/. However, other skills in this repo do use shell wrappers (e.g., e2e-health/scripts/, merge-queue/scripts/), and the wrapper provides a simpler CLI for rotating forum hosts. The SKILL.md already includes Bash(python3 .../gather.py:*), making it redundant for agent usage.

  • [prompt-injection] skills/user-forum-whats-new/scripts/gather.py — Release bodies and PR titles fetched from the GitHub API flow directly into JSON output consumed as LLM context. Repos are hardcoded to first-party orgs (fullsend-ai), limiting exposure. Consistent with known threat surface rather than a novel gap introduced by this PR.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (2)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes.

Low

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py — No test covers draft release filtering. The production code in classify() skips releases with draft: True, but no test verifies this behavior. A regression removing the draft-skip guard could cause draft releases to set the cutoff timestamp or appear in the output.

  • [scope-vs-convention] skills/user-forum-whats-new/SKILL.md — AGENTS.md line 22 conditionally requires checking user-facing guides for cross-reference opportunities when adding a new skill. The PR body explicitly states this is "intentionally not cross-linked" because it's a forum-host skill. The instruction is conditional ("if the new skill fills a gap in those guides"), and peer contributor skills (topissues, nextwork) are not cross-referenced in those guides either. Clarifying this exemption in AGENTS.md would prevent recurring review findings on contributor-tooling skills.

  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py — The script uses sys.stderr.write() for error/warning output (6 call sites outside the two early guards) and raise SystemExit() for the entry point, while both peer skill scripts (topissues.py, nextwork.py) uniformly use print(..., file=sys.stderr) and sys.exit() respectively. The entry point uses raise SystemExit(main()) with main() -> int, whereas both peers use bare main() with main() -> None. The two early guard clauses (Python version check and timezone check) have a legitimate reason for sys.stderr.write/raise SystemExit since they run before 3.11-dependent imports.

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — The shell wrapper is a novel sub-pattern: a thin .sh wrapper solely to exec a .py sibling has no precedent in skills/. Peer Python-based skills invoke their Python scripts directly from SKILL.md allowed-tools. The SKILL.md already includes Bash(python3 .../gather.py:*), making the wrapper redundant for agent usage.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (3)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes.

Low

  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py — The script uses sys.stderr.write() for error/warning output (6 call sites) and raise SystemExit() for exits (3 call sites), while both peer skill scripts (topissues.py, nextwork.py) uniformly use print(..., file=sys.stderr) and sys.exit() respectively. The entry point uses raise SystemExit(main()) with main() -> int, whereas both peers use bare main() with main() -> None. The two early guard clauses (Python version check and timezone check) have a legitimate reason for sys.stderr.write/raise SystemExit since they run before 3.11-dependent imports.

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py — No test covers draft release filtering. The production code in classify() skips releases with draft: True, but no test verifies this behavior. A regression removing the draft-skip guard could cause draft releases to set the cutoff timestamp.

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md — The allowed-tools frontmatter grants Bash(xdg-open:*) and Bash(open:*) which accept arbitrary arguments; Bash(gh search:*) grants all search subcommands when only gh search prs is used. These are broader than necessary for the documented use cases. Per ADR 0027, allowed-tools is a steering mechanism, not a security boundary.

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — The shell wrapper is a novel sub-pattern: while shell scripts in skills/*/scripts/ are common, a thin .sh wrapper solely to exec a .py sibling has no precedent. Peer Python-based skills invoke their Python scripts directly from SKILL.md allowed-tools.

  • [missing-authorization] — Non-trivial addition (~1175 lines) with no linked issue. The PR body states "intentionally untracked forum-host tooling." A tracking issue would provide discoverability and design discussion space.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (4)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes.

Medium

  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py — The script uses sys.stderr.write() for error/warning output and raise SystemExit() for exits, while the established pattern in peer skills (topissues.py, nextwork.py) uses print(..., file=sys.stderr) and sys.exit() respectively. The entry point uses raise SystemExit(main()) with main() -> int, whereas peers use bare main() with main() -> None. The two early guard clauses (Python version check and timezone check) have a legitimate reason for sys.stderr.write/raise SystemExit since they run before 3.11-dependent imports.
    Remediation: Replace sys.stderr.write(...) with print(..., file=sys.stderr) and raise SystemExit(...) with sys.exit(...) in post-import code. Change the entry point pattern to match peers.

  • [architecture-cross-reference] skills/user-forum-whats-new/SKILL.md — AGENTS.md line 21 requires checking user-facing guides for cross-reference opportunities when adding a new skill under skills/. The PR body declines, citing "forum-host skill, not a sandbox agent skill." While the instruction is conditional ("if the new skill fills a gap in those guides"), AGENTS.md does not define an exception for contributor-oriented skills.
    Remediation: Verify the check was performed and note in the PR that no relevant cross-reference opportunity exists, or add a brief mention in docs/guides/user/customizing-with-skills.md.

Low

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py — No test covers draft release filtering. The production code in classify() skips releases with draft: True, but no test verifies this behavior. A regression removing the draft-skip guard could cause draft releases to set the cutoff timestamp.

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — The shell wrapper is a novel sub-pattern: while shell scripts in skills/*/scripts/ are common, a thin .sh wrapper solely to exec a .py sibling has no precedent. Peer Python-based skills invoke their Python scripts directly from SKILL.md allowed-tools.

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md — The allowed-tools frontmatter grants permissions slightly broader than documented use cases: Bash(xdg-open:*) and Bash(open:*) accept arbitrary arguments (Step 6 documents the /tmp/ use case); Bash(gh search:*) grants all search subcommands when only gh search prs is used; WebFetch is granted but not used by the skill's scripts. Per ADR 0027, allowed-tools is a steering mechanism, not a security boundary.

  • [missing-authorization] — Non-trivial addition (~1143 lines) with no linked issue. The PR body states "intentionally untracked forum-host tooling." A tracking issue would provide discoverability and design discussion space.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (5)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/ — This PR adds 4 files under skills/, which is a protected path requiring human approval. The affected files are skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, and skills/user-forum-whats-new/scripts/gather_test.py. The PR has no linked issue to authorize the change to governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Obtain explicit human maintainer approval for the protected-path changes.

Medium

  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py — The script uses sys.stderr.write() for all error/warning output (6 call sites) and raise SystemExit() for exits (3 call sites), while the established pattern in skills/topissues/scripts/topissues.py and skills/nextwork/scripts/nextwork.py uses print(..., file=sys.stderr) and sys.exit() respectively. The existing scripts consistently use these patterns across dozens of call sites.
    Remediation: Replace sys.stderr.write(...) with print(..., file=sys.stderr) and raise SystemExit(...) with sys.exit(...) to match codebase conventions.

Low

  • [test-inadequate] skills/user-forum-whats-new/scripts/gather_test.py — No test covers draft release filtering. The production code in classify() skips releases with draft: True, but no test verifies this behavior. A draft release with a recent published_at could incorrectly set the cutoff if the draft-skip logic were removed or broken.

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — The shell wrapper is a novel pattern: no other skill uses a .sh wrapper around a .py script. Existing skills (topissues, nextwork) invoke Python scripts directly. The allowed-tools already grants Bash(python3 skills/user-forum-whats-new/scripts/gather.py:*), making the wrapper redundant for Claude Code usage. If kept for human convenience, consider documenting why.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (6)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, skills/user-forum-whats-new/scripts/gather_test.py — Four files under the skills/ protected path are modified. The PR has no linked issue providing justification for modifying governance/infrastructure files. Human approval is required for all protected-path changes regardless of context.
    Remediation: Link a tracking issue or obtain explicit human approval for the protected-path changes.

Medium

  • [permission-scope-open] skills/user-forum-whats-new/SKILL.md:8 — The allowed-tools frontmatter grants Bash(xdg-open:*) and Bash(open:*) with unrestricted glob patterns. These commands accept arbitrary URLs and file paths, but the skill only needs to open the generated HTML at /tmp/fullsend-whats-new-YYYY-MM-DD.html. An LLM agent could be prompt-injected (via crafted release bodies or PR titles fetched from the API) into opening arbitrary URLs or invoking protocol handlers.
    Remediation: Restrict to Bash(xdg-open /tmp/fullsend-whats-new-*:*) and Bash(open /tmp/fullsend-whats-new-*:*), or remove and instruct the user to open the file manually.

Low

  • [permission-scope-gh] skills/user-forum-whats-new/SKILL.md:8 — Direct gh CLI permissions (gh search, gh pr view, gh pr diff, gh pr list, gh issue view, gh issue list, gh release view, gh release list) are granted beyond what gather.py encapsulates. These are read-only subcommands justified by the example-hunting workflow (Step 3), consistent with cutting-releases precedent.

  • [missing-authorization] — No linked issue. The PR body documents rationale ("intentionally untracked forum-host tooling"). No AGENTS.md policy requires linked issues for all PRs, but the convention helps discoverability.

  • [redundant-computation] skills/user-forum-whats-new/scripts/gather.py — window_bounds() is computed twice with identical arguments: in main() and again inside build_output(). Currently safe because the same now is threaded through, but fragile if a future caller passes different values.

  • [pattern-inconsistency] skills/user-forum-whats-new/SKILL.md:8 — allowed-tools grants both Bash(bash .../gather.sh:*) and Bash(python3 .../gather.py:*). Since gather.sh is a thin exec python3 wrapper, both paths are redundant. Peer skills pick one invocation path.

  • [docstring-convention] skills/user-forum-whats-new/scripts/gather.py:2 — Peer scripts (topissues.py, nextwork.py) include (stdlib only) in module docstrings. This marker is missing from gather.py.

  • [edge-case] skills/user-forum-whats-new/scripts/gather.py — End-of-day boundary time(23, 59, 59, 999999) is practically equivalent to midnight for GitHub's second-granularity timestamps.

  • [indirect-prompt-injection] skills/user-forum-whats-new/scripts/gather.py — Release bodies and PR titles from the GitHub API flow verbatim into LLM context. Mitigated by hardcoded trusted-org repo scope (fullsend-ai/fullsend and fullsend-ai/agents).


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (7)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, skills/user-forum-whats-new/scripts/gather_test.py — This PR modifies 4 files under the protected skills/ path. The PR has no linked issue, so the rationale for modifying governance/infrastructure files cannot be verified against authorized work. Human approval is always required for protected-path changes.
    Remediation: Link a tracking issue that authorizes the skill addition, or obtain explicit human approval for the protected-path changes.

Medium

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md:9 — The allowed-tools frontmatter grants the Write tool, which no other skill in this repository requests. This gives the skill agent unscoped filesystem write capability. Other skills with output needs (topissues, nextwork) rely on stdout. The skill workflow (step 6) instructs writing to /tmp/fullsend-whats-new-YYYY-MM-DD.html, which explains the need, but the permission is broader than necessary.
    Remediation: Document the Write tool justification explicitly in the SKILL.md, or consider using stdout output matching the topissues/nextwork pattern.

Low

  • [code-organization] skills/user-forum-whats-new/scripts/gather.sh — No other Python-based skill uses an intermediary shell wrapper. Both nextwork and topissues invoke python3 directly via allowed-tools. The wrapper is redundant since the SKILL.md already includes Bash(python3 .../gather.py:*) in allowed-tools.
  • [error-handling-idiom] skills/user-forum-whats-new/scripts/gather.py:22 — Uses raise SystemExit(1) (lines 22 and 36) instead of sys.exit(1). All other Python scripts in this repo use sys.exit() consistently. While functionally identical, this diverges from the established convention.
  • [missing-authorization] This is a non-trivial change (~967 lines) with no linked issue. The PR body states this is "intentionally untracked forum-host tooling," which is plausible but informal.
  • [cross-reference-gap] skills/user-forum-whats-new/SKILL.md — AGENTS.md instructs checking user-facing guides for cross-reference opportunities when adding a skill. The PR opts out, arguing this is a forum-host skill, not a sandbox agent skill. This is defensible — the skill does not fill a gap in those guides.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (8)

Review

Findings

Medium

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, skills/user-forum-whats-new/scripts/gather_test.py — All new files are under skills/, which is a protected path requiring human approval. The PR links to issue Resolve skill definition inconsistency between released version and main branch #6353 and explains the rationale. Human approval is always required for protected-path changes, regardless of context.

  • [issue-mismatch] PR claims “Fixes Resolve skill definition inconsistency between released version and main branch #6353” but the issue’s acceptance criteria focus on reconciling an existing diverged skill definition (identify the diverged skill, reconcile released/main versions, verify consistency, consider a CI check). This PR adds a brand new skill that has never been released, so there is no “drift” to reconcile. It is plausible the author believes the “drift” was that this skill existed internally but was missing from main.
    Remediation: Clarify in the PR body how this skill addresses the specific acceptance criteria in Resolve skill definition inconsistency between released version and main branch #6353, or unlink Resolve skill definition inconsistency between released version and main branch #6353 and reference an appropriate issue.

  • [commit-convention] PR title uses chore(skills): prefix but adds 600+ lines of new functionality (skill definition, Python scripts, tests, Makefile integration). Per COMMITS.md, chore is for maintenance (CI, deps, tooling). However, COMMITS.md also says “prefer refactor or chore over feat” and this is a contributor skill, not a user-facing feature.
    Remediation: Consider whether feat(skills): or chore(skills): better reflects whether this should appear in release notes.

  • [architectural-deviation] skills/user-forum-whats-new/SKILL.md — AGENTS.md requires checking user-facing guides for cross-reference opportunities when adding a new skill. PR body explicitly opts out, citing “this is a forum-host skill, not a sandbox agent skill.” The AGENTS.md qualifier “if the new skill fills a gap in those guides” makes the requirement conditional, and the author’s reasoning is substantive even if the formal exemption is undocumented.
    Remediation: Either add a brief cross-reference to docs/guides/user/customizing-with-skills.md, or document the exemption criteria for contributor-oriented skills.

Low

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md — The allowed-tools entries Bash(open:*) and Bash(xdg-open:*) permit executing these commands with arbitrary arguments. The intent is to open a local HTML file in /tmp/, but the wildcard allows any argument. Runs in a local developer session with limited blast radius.
    Remediation: Constrain to specific file patterns, e.g. Bash(open /tmp/fullsend-whats-new-*:*) and Bash(xdg-open /tmp/fullsend-whats-new-*:*).

  • [permission-expansion] skills/user-forum-whats-new/SKILL.md — Bash(gh api repos/:*) allows API calls to any repo path, not just the two repos the skill operates on (fullsend-ai/fullsend and fullsend-ai/agents). Combined with Bash(gh search:*), this grants the skill access beyond its stated scope. Consistent with existing codebase conventions.

  • [edge-case] skills/user-forum-whats-new/scripts/gather.py:137 — PR field closedAt is used as the merge timestamp. For merged PRs found via gh search prs --merged, closedAt equals mergedAt in practice; the search endpoint does not expose mergedAt directly.

Previous run (9)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.py, skills/user-forum-whats-new/scripts/gather.sh, skills/user-forum-whats-new/scripts/gather_test.py — All new files are under skills/, which is a protected path requiring human approval. The PR has no linked issue providing authorization for modifying governance/infrastructure files.
    Remediation: Link an issue authorizing the addition of this skill, or obtain explicit human reviewer approval for the protected-path changes.

Low

  • [logic] skills/user-forum-whats-new/scripts/gather.py:257 — In main(), window_bounds(since, until) is called twice with different implicit now values: once at line 257 for validation/fetch and once inside build_output() at line 201 (via the build_output → window_bounds call chain). When --until is today, the second call computes a slightly later until_ts. The fetched data is bounded by the first (narrower) window, so no unfetched data appears in output. The only consequence is cosmetic: window_end_utc in JSON output reflects the second now, not the actual search boundary used by gh.
    Remediation: Capture now=datetime.now(UTC) once in main() and pass it through to build_output().

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (10)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.sh — Both new files are under skills/, which is a protected path requiring human approval. The PR has no linked issue providing authorization for modifying governance/infrastructure files.
    Remediation: Link an issue authorizing the addition of this skill, or obtain explicit human reviewer approval for the protected-path changes.

Low

  • [error handling] skills/user-forum-whats-new/scripts/gather.sh:100 — The Python load() function catches all exceptions and returns an empty fallback list. If gh api or gh search writes non-JSON output while still exiting 0, load() silently returns [] with no warning. The primary failure mode (non-zero exit from gh) is already handled by set -euo pipefail, which aborts the script before Python runs, so the practical impact is limited to edge cases.
    Remediation: Add a sys.stderr.write(f'warning: failed to parse {name}: {e}\n') in the except block of load() so the caller knows data was dropped.

  • [missing-authorization] skills/user-forum-whats-new/SKILL.md — Non-trivial new skill (~285-line SKILL.md + ~216-line script) added without a linked issue. The PR body states "N/A — forum-host tooling" but there is no issue documenting the need or authorization for this work. See also: [protected-path] finding for the same files.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (11)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.sh — Both new files are under skills/, which is a protected path requiring human approval. The PR has no linked issue providing authorization for modifying governance/infrastructure files.
    Remediation: Link an issue authorizing the addition of this skill, or obtain explicit human reviewer approval for the protected-path changes.

Low

  • [error handling] skills/user-forum-whats-new/scripts/gather.sh:105 — The Python load() function catches all exceptions and returns an empty fallback list. If gh api or gh search writes non-JSON output while still exiting 0, load() silently returns [] with no warning. The primary failure mode (non-zero exit from gh) is already handled by set -euo pipefail, which aborts the script before Python runs, so the practical impact is limited to edge cases.

  • [missing-authorization] skills/user-forum-whats-new/SKILL.md — Non-trivial new skill (266 lines + 186-line script) added without a linked issue. The PR body states "N/A — forum-host tooling" but there is no issue documenting the need. See also: [protected-path] finding for the same files.

  • [scope-cross-reference-gap] skills/user-forum-whats-new/SKILL.md — AGENTS.md line 21 requires checking user-facing guides when adding a new skill under skills/. The PR body explicitly addresses this ("forum-host skill, not a sandbox agent skill") and the reasoning is sound, but AGENTS.md does not carve out an exception for non-user-facing skills.


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (12)

Review

Findings

High

  • [protected-path] skills/user-forum-whats-new/SKILL.md, skills/user-forum-whats-new/scripts/gather.sh — Both new files are under skills/, which is a protected path requiring human approval. The PR has no linked issue providing authorization for modifying governance/infrastructure files.
    Remediation: Link an issue authorizing the addition of this skill, or obtain explicit human reviewer approval for the protected-path changes.

Low

  • [error handling] skills/user-forum-whats-new/scripts/gather.sh:11 — When --since or --until is the last argument with no value, ${2:-} assigns an empty string, then shift 2 exits the script under set -e with no diagnostic output instead of the friendly error message on line 25. Consider guarding with an argument count check before accessing $2.

  • [missing-authorization] skills/user-forum-whats-new/SKILL.md — Non-trivial new skill (263 lines + script) added without a linked issue. The PR body states "N/A — forum-host tooling" but there is no established precedent for how individual post-launch skills should be added to this repo. Consider filing an issue to document the need.

  • [documentation-alignment] skills/user-forum-whats-new/SKILL.md — AGENTS.md line 21 states to check user-facing guides for cross-reference opportunities when adding skills under skills/. The PR intentionally skips this, stating it's "forum-host tooling, not a sandbox agent skill." The guidance is ambiguous about whether repo-root contributor skills require the same cross-linking as user-facing agent skills. Consider clarifying AGENTS.md or adding a note to the docs explaining the distinction.


Labels: PR adds a new contributor skill under skills/


Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review fullsend-ai-review Bot added component/skills type/chore Maintenance and housekeeping tasks labels Aug 18, 2026
@ascerra

ascerra commented Aug 18, 2026

Copy link
Copy Markdown
Contributor Author

Re: protected-path finding — this PR intentionally adds a new contributor skill under skills/ (forum-host tooling, not sandbox agent config). It follows the same layout as skills/topissues, skills/filing-issues, etc. No linked issue because it is net-new host documentation + a gather helper, not a change to existing governance. Happy to link a tracking issue if maintainers want one before merge.

@fullsend-ai-review

fullsend-ai-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

🤖 Review · ❌ Terminated · Started 11:38 AM UTC · Ended 11:53 AM UTC

Commit: 3cdd500 · View workflow run →

@rh-hemartin

Copy link
Copy Markdown
Member

This sounds as reease notes to me, should we generalize this to "from X date to Y date", "from X commit to Y commit"?

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:38 AM UTC · Completed 11:53 AM UTC

Commit: 3cdd500 · View workflow run →

@fullsend-ai-review

fullsend-ai-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 12:56 PM UTC · Completed 1:11 PM UTC

Commit: 5b43768 · View workflow run →

fullsend-ai-review[bot]

This comment was marked as outdated.

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Review-only sweep: 7 inline findings (1 high, 6 medium) on the gather script and SKILL.md. No approval or change request implied.

Comment thread skills/user-forum-whats-new/SKILL.md Resolved
Comment thread skills/user-forum-whats-new/SKILL.md Resolved
@ascerra

ascerra commented Aug 18, 2026

Copy link
Copy Markdown
Contributor Author

@rh-hemartin Keeping this skill scoped to the user-forum What's New talk track on purpose — it is deliberately not release notes (ranking, talkable filters, clickable examples, Released vs On main for a 4–6 minute verbal pass).

gather.py already accepts arbitrary --since / --until dates for that workflow. Commit-to-commit gathering would be a different fetch model (compare / git log) and is a reasonable follow-up if someone wants a separate “what landed between refs” helper — not this PR's job.

@fullsend-ai-review

fullsend-ai-review Bot commented Aug 18, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:01 PM UTC · Completed 11:18 PM UTC

Commit: 436dd1f · View workflow run →

fullsend-ai-review[bot]

This comment was marked as outdated.

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Review-only sweep: 3 inline findings (3 medium) on gather.py and SKILL.md frontmatter. No approval or change request implied.

Comment thread skills/user-forum-whats-new/SKILL.md Resolved
@ascerra
ascerra force-pushed the chore/skills-user-forum-whats-new branch from 436dd1f to 6a0cfb0 Compare August 19, 2026 19:19
@fullsend-ai-review

fullsend-ai-review Bot commented Aug 19, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 7:21 PM UTC · Completed 7:39 PM UTC

Commit: 6a0cfb0 · View workflow run →

@fullsend-ai-review
fullsend-ai-review Bot dismissed stale reviews from themself August 19, 2026 19:39

Superseded by updated review

@fullsend-ai-review fullsend-ai-review Bot added the requires-manual-review Review requires human judgment label Aug 19, 2026

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Review-only sweep: 1 high finding in the PR description (below) plus 2 medium inline findings on SKILL.md. No approval or change request implied.


[HIGH] Fixes #6353 links an unrelated open issue and will auto-close it on merge

The PR body still reads Fixes #6353 at head 6a0cfb07 (re-verified against the PR body via the API). That issue is "Resolve skill definition inconsistency between released version and main branch" — an open, scribe-generated issue about drift between the released (deployed) agent skill definition and its main-branch counterpart, with acceptance criteria "Identify the specific skill definition that has diverged", "Reconcile the released version and main branch version", and adding a CI drift check.

This PR adds a brand-new forum-host recap skill and addresses none of those criteria, so merging would silently close unrelated tracked work.

Context that makes this look like a mis-fill rather than a deliberate link: the review bot raised a protected-path finding about the missing linked issue, the reply was "No linked issue because it is net-new host documentation + a gather helper... Happy to link a tracking issue if maintainers want one", and an unrelated issue number appeared in the body afterwards. I also searched open and closed issues for a forum-recap / user-forum-skill tracker (gh search issues --repo fullsend-ai/fullsend "forum What's New recap" and "user forum skill") and found none, so there is no correct issue to swap in.

Suggested fix: Drop the Fixes #6353 line, or downgrade it to a non-closing reference (Refs #6353) only if there is a real relationship. Then either file a tracking issue for the forum recap skill and link that, or state in the body that the work is intentionally untracked (matching the earlier reply to the protected-path finding).

Comment thread skills/user-forum-whats-new/SKILL.md Resolved
Comment thread skills/user-forum-whats-new/SKILL.md Resolved
ascerra and others added 7 commits September 9, 2026 10:58
Require UTF-8 list JSON from fetch outputs (non-zero exit on corrupt
payloads), filter gh search to --base main, and treat #forum-fullsend-ai
as host-supplied input.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…e claim

Emit second-precision Z timestamps for gh search --merged-at, and require
running gather from a fullsend checkout so allowed-tools path grants match.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Chai claimed these fixes without pushing. Add --slurp (+ flatten) for
paginated releases, emit until_clamped for future --until clamps, and
drop the non-existent timezone-data package name.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Reject ISO week-dates, add gh subprocess timeouts, and use a half-open
lower bound so consecutive Tuesday windows do not overlap.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Hosts rotate and run gather fresh locally — do not chain window_end_utc.
Correct the half-open overclaim and reword to_z docstring to describe our
emit choice rather than unverified GitHub rejection of fractional seconds.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop the public auth-walled Google Doc URL, monkeypatch fetch_into
subprocess argv, and error clearly when the window has not started.

Signed-off-by: Adam Scerra <ascerra@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Signed-off-by: Adam Scerra <ascerra@redhat.com>
Signed-off-by: Adam Scerra <ascerra@redhat.com>
@fullsend-ai-review

fullsend-ai-review Bot commented Sep 9, 2026 •

Copy link
Copy Markdown

🤖 Review · ⚠️ Cancelled · Started 3:01 PM UTC · Ended 3:02 PM UTC

Commit: dd41a9f · View workflow run →

@fullsend-ai-review

fullsend-ai-review Bot commented Sep 9, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 3:03 PM UTC · Completed 3:18 PM UTC

Commit: 0ed9843 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5 · Effort: high · Cost: $2.90

@fullsend-ai-review fullsend-ai-review 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.

See the review comment for full details.

Comment thread skills/user-forum-whats-new/SKILL.md Resolved

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Re-review at 0ed98435. I re-checked every finding I raised across the earlier rounds against the code at head rather than relying on the resolved marks, and ran the skill for real. All of them hold up. No medium-or-higher findings this round; one non-blocking low inline.

Verified at head:

Earlier finding State at 0ed98435
gh search prs had no --base filter fixed — gather.py:278-279
gh api --paginate without --slurp produced concatenated arrays fixed — --slurp at gather.py:260 plus flatten_gh_slurp() (134) which also handles the flat and empty-page shapes
no timeout on gh subprocesses fixed — GH_TIMEOUT_SEC (43) applied to both calls (265, 291), mapped to exit 1 in main (386)
one cross-repo release cutoff fixed — per-repo cutoffs (209-210), surfaced as the release_cutoff_utc map
prereleases could set the cutoff fixed — prereleases stay as candidates but never enter times (191-196)
inclusive window boundary double-counted the shared 08:00 ET instant fixed — half-open in in_window (131); the residual late-run band is documented in window_bounds (101-104) and SKILL.md, with host dedupe as the mitigation
clamp-to-now inverted the window with a misleading two-identical-dates error fixed — separate "window has not started yet" (113) and "until before since" (117) paths
search-truncation warning was stderr-only fixed — search_truncated is now in the JSON (321, 326) as well as stderr
Step 6 assembled HTML from untrusted PR titles with no escaping requirement fixed — explicit escaping constraint at SKILL.md:255-257
opener grants carried a dead :* suffix fixed. Re-checked against the current permissions doc: * "stands in for whatever text is in its place", and the space-before-* rule that broke the old form does not apply here because the * is attached to the prefix (the doc's own Bash(ls*) matches lsof case). Bash(xdg-open /tmp/fullsend-whats-new-*) does match xdg-open /tmp/fullsend-whats-new-2026-09-09.html.
~130 lines of Python in a heredoc escaped every lint and test gate fixed, and the gate is real, not nominal: ruff.toml selects E,F,I,W,UP,B,SIM over ., Makefile:201 runs gather_test.py under script-test, and .github/workflows/lint.yml:58 runs make script-test. The # noqa: UP036 on the version guard (19) is consistent with target-version = "py312".
fetch_into was "tested" by grepping its own source fixed — test_fetch_into_argv_and_timeout patches gather.subprocess.run and asserts argv and timeout for real

Executed checks:

  • python3 skills/user-forum-whats-new/scripts/gather_test.py — 34 tests, OK.
  • Live end-to-end run: bash skills/user-forum-whats-new/scripts/gather.sh --since 2026-09-02 --until 2026-09-09 — exit 0, clean stderr. Window resolved to 2026-09-02T12:00:00Z .. 2026-09-09T15:30:47Z with until_clamped: true, which is the correct 08:00 EDT to now mapping. Per-repo cutoffs came back distinct (fullsend 20:39:34Z, agents 20:39:54Z), and the split was 54 released / 48 on main, with the boundary landing where the v0.42.0 publish times say it should.
  • Sanity-checked the classification heuristic against real data: the last 25 fullsend releases and last 10 agents releases are all target_commitish=main, non-draft and non-prerelease. So the max-in-window-publish-time cutoff is sound for this repo's actual cadence, and the draft/prerelease branches are defensive rather than load-bearing. Worth knowing that the one scenario that would mislabel — a patch release cut from a non-main branch mid-window — has no precedent in either repo's history, so I am not raising it.

The two open threads are both bot lows. The draft-release one is accurate: classify() skips draft releases at line 186 and no test covers that branch — it is the only untested branch in the function, and defensive-only given the data above.

Comment thread skills/user-forum-whats-new/SKILL.md Resolved

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Approving at 0ed98435.

Every finding I raised across the earlier rounds is verified fixed at this head — details and the executed checks are in my re-review comment on this commit (34/34 unit tests, a clean live end-to-end gather.sh run, and confirmation that make script-test actually runs gather_test.py in CI).

The one inline item I left is an explicitly non-blocking low about scoping the Write grant; take it or soften the prose, either is fine by me. Thanks for the patience across the rounds — the permission scoping and the window math both ended up in good shape.

@ascerra
ascerra added this pull request to the merge queue Sep 9, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Sep 9, 2026
@ascerra
ascerra added this pull request to the merge queue Sep 9, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Sep 9, 2026
@rh-hemartin
rh-hemartin added this pull request to the merge queue Sep 10, 2026
Merged via the queue into main with commit 90353cb Sep 10, 2026
72 checks passed
@rh-hemartin
rh-hemartin deleted the chore/skills-user-forum-whats-new branch September 10, 2026 08:02
@fullsend-ai-retro

fullsend-ai-retro Bot commented Sep 10, 2026 •

Copy link
Copy Markdown

🤖 Finished Retro · ✅ Success · Started 8:04 AM UTC · Completed 8:22 AM UTC

Commit: 0ed9843 · View workflow run →

Runtime: claude · Model: sonnet → claude-sonnet-5 · Effort: high · Cost: $3.08

@fullsend-ai-retro

Copy link
Copy Markdown

PR #6314 (user-forum-whats-new skill) took ~22 days and 13 review-agent rounds to merge, but the delay was almost entirely human-availability silence (two gaps of 7 and 6.6 days), not agent turnaround (every review/fix run completed in 11-20 minutes). The review agent (fullsend-ai-review) was reliable at governance/policy checks (protected-path, permission-scope, commit-convention) but missed roughly 20 substantive correctness/security bugs across 13 rounds - GitHub CLI pagination/quoting bugs, timezone/window-boundary math, silent-failure masking, a hardcoded internal Google Doc URL, fake self-referential tests, missing subprocess timeouts, missing HTML-escaping for untrusted PR titles - all caught instead by human reviewer waynesun09 across 8 review sweeps, including a live end-to-end execution of the tool before approving. Root-cause tracing (via the agents repo discovered from workflow logs, fullsend-ai/agents) shows the correctness/security/style/docs sub-agents dispatched by the pr-review skill have tools: Read, Grep, Glob only - no Bash - so they cannot execute code, tests, or --help checks; only the risk-assessment sub-agent gets a narrowly scoped Bash grant. This is filed as a new proposal below since I found no existing issue covering it, and it also blocks the fix already proposed in open issue #778 (which assumes Bash access the sub-agent doesn't have).

Evidence found for several already-open issues in fullsend-ai/agents / fullsend-ai/fullsend (no new issues filed for these - see linked issues):

  • agents#552 (fix agent should prohibit git commit -s/Signed-off-by, still open): the PR's one /fs-fix invocation (2026-09-02) failed exactly this way, wasting ~$3.58; the strip-and-continue fix (Follow-up from PR #1182: The --mint-url flag row shows no default, but it now defaults to the hosted mint URL.... #1184) merged 3 days later (2026-09-05), so this run just missed it. Corroborates post-review.sh does not apply outcome labels (ready-for-merge, requires-manual-review) #552's premise that the model-level prohibition still needs strengthening.
  • agents#106 (review agent re-raises findings the author explicitly declined, still open): the same pattern recurred here - an AGENTS.md cross-reference finding was pushed back on by the author in round 1 but kept resurfacing through round 7, where the agent's own language finally softened to 'this is defensible' without any code change prompting it.
  • agents#778 (correctness sub-agent should verify external CLI tool behavior, still open): this PR's gh search --limit 100 truncation and gh api --paginate without --slurp bugs are additional real-world instances of exactly the gap Prioritize workflows should not be installed in .fullsend config repo #778 describes.
  • agents#1135 (review agent should detect circular test confidence, still open): this PR had a related instance - a 'test' that grepped its own function's source via inspect.getsource instead of executing it, caught only by the human.
  • fullsend#5004 (review agent should flag silent-failure-mode degradation, still open): this PR's load_json fail-open behavior (bad JSON from the tool's own output silently producing an empty-but-valid recap) is the same class of gap.
  • fullsend#3412 (chai-bot functionality expansion gated on pending security review, still open): a third-party bot (redhat-chai-bot) posted three confident 'Fixed ✅' replies on this PR describing patches that were never actually committed; the author caught the discrepancy manually. This is concrete, timely evidence for that pending security review.

Overall the human review (waynesun09's 8 sweeps plus a live tool execution before approving) was substantially more rigorous than the agent review for this PR's domain (timezone/API/pagination-heavy scripting) - this isn't a criticism of any individual approval, just a signal that this class of change currently needs a human safety net the agent can't yet provide.

Proposals filed

This branch was previously deployed

1 inactive deployment
dev — 0ed98435 Deployed Sep 9, 2026 by ascerra via behaviour #12399
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

component/skills risk/elevated PR risk: elevated type/chore Maintenance and housekeeping tasks

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants