Skip to content

chore(agents): align instructions and skills - #2252

Merged
Benoît Cortier (CBenoit) merged 4 commits into
masterfrom
codex/agent-instructions-maintenance
Oct 27, 2026
Merged

Benoît Cortier (CBenoit) merged 4 commits into
masterfrom
codex/agent-instructions-maintenance

Conversation

@CBenoit

@CBenoit Benoît Cortier (CBenoit) commented Oct 27, 2026 •

Copy link
Copy Markdown
Member

Keep repository guidance accurate and concise, refresh shared skills from Gateway, and expose every skill through an individual Claude symlink.
Omit redundant opening titles and synchronize the authoring and review rules that prevent them.
Remove redundant skill-discovery and synchronization boilerplate from the root instructions.

Benoît Cortier (CBenoit) commented Oct 27, 2026 •

Copy link
Copy Markdown
Member Author

Validation: all 34 configured skill directories match Gateway's recursive file sets and SHA-256 hashes. All 3 AGENTS/CLAUDE pairs and 42 individual skill links resolve correctly; .claude/skills is a real directory. No local settings files remain.

All 8 changed repository-local skills passed Tessl and the skill-creator validator; 66 unique canonical skills passed both across the eight repositories. Independent review, exact title-only comparisons, frontmatter checks, and git diff --check pass. The rebased heading/instruction patches are unchanged by range-diff; the final two-file shared-test refresh was independently verified byte-for-byte against Gateway.

No application or intent files are changed by this PR. The sspi-rs and IronRDP branches now include their merged compile fixes through the updated base.

Lines changed: production code +0/−0; tests +0/−0; other files +148/−232 across 91 files.

Intent naming consistency: all eight root AGENTS files cover shared parts, all eight intent-files skills match Gateway byte-for-byte, and the Gateway/IronVNC documentation copies are identical. Independent GPT-6.1 Sol High review and git diff --check passed; all 420 tracked symlinks remain symbolic links.

Note

Human-tuned, LLM-assisted content.

@github-actions github-actions Bot added automation-failed Exact-head automated classification or review failed or was unavailable risk/unknown Risk could not be determined automatically; needs maintainer-level scrutiny scope/tooling Build, CI, release, or developer tooling size/XXL Size: 1300 or more counted lines or 50 or more files labels Oct 27, 2026

Benoît Cortier (CBenoit) commented Oct 27, 2026 •

Copy link
Copy Markdown
Member Author

CI note: Checks [linux] fails while compiling an existing EGFX test. crates/ironrdp-egfx/src/client.rs:3404 supplies four arguments to progressive_tile_stream, whose declaration at line 2756 takes three (E0061).

The source file, root manifest, lockfile, and CI workflow have identical Git blobs in this PR and its base 3ca12ee63. This maintenance PR changes no Rust source. The failure is documented without expanding the instruction/skill changes.

The compile fix is proposed separately in #2253; merge that fix before re-running this PR's checks.

Note

LLM-assisted content (no human feedback).

@github-actions github-actions Bot added the needs-author-action The pull request author is the current next actor label Oct 27, 2026

Benoît Cortier (CBenoit) commented Oct 27, 2026 •

Copy link
Copy Markdown
Member Author

Instruction change rationale — reviewed diff at f65ac08f2.

File / instruction block Addition, deletion, or replacement Rationale / where the guidance remains
AGENTS.md Use INTENT*.md and .intent.md in the intent-file rule. Includes the documented INTENT..md and .intent..md shared parts; ownership and read-before-edit obligations are unchanged.
3 AGENTS files Remove 3 opening repository/path/topic banners. Their scope or purpose is already clear from the location and retained body; these titles organize no distinct sections. The instructions themselves remain.
42 skill opening titles Remove opening titles that repeat skill names or frontmatter descriptions. The metadata already supplies the identity and purpose. Title-only edits preserve frontmatter, the remaining body, code blocks, and organizing headings.
Shared authoring/review rules Mirror the three prevention rules from Gateway unchanged. Keeps agents-md, agents-md-review, and skill-discipline aligned with the authoritative source rather than creating local variants.
Shared test guidance Refresh test-oracle-reviewer and test-pruner from latest Gateway master. Those upstream files changed while these PRs were being rebased. Exact copies restore source parity; their bodies were not independently customized or semantically re-audited.
AGENTS.md Delete the two skills-discovery / weekly-sync openers. Explicit maintainer request: these lines duplicate harness discovery and sync configuration rather than repository policy.
AGENTS.md Collapse three generated-changelog sentences into one release-only rule. The same prohibition was repeated three times. The retained sentence still bars feature/fix work from editing generated crate changelogs.
AGENTS.md Move the PDU error / unknown-value STYLE pointer into protocol-reviewer. That guidance applies to PDU changes rather than every task. It remains in the reviewer that needs it, instead of keeping a task-specific guide pointer always loaded.
AGENTS.md Remove the protocol-skill installation command while retaining the windows-protocols pointer. Installation syntax is discoverable tooling detail. The repository-specific choice of specification navigator remains.
.agents/skills/commit-scope/SKILL.md Add existing channel/backend scopes and clarify native, UDP, USB, and replay ownership. The current workspace contains these components; the older list omitted them or described only part of their ownership. The existing one-owner and test-scope rules are preserved.
.agents/skills/crate-readme-writer/SKILL.md Delete generic writing checklist items, the “write README directly” command, and the background attribution. docs-writer already supplies prose rules, and the command/background adds no task constraint. README-before-implementation, its reason, repository fit, integration contract, scope feedback, and verification remain.
.agents/skills/new-crate/SKILL.md Combine cargo-new and the no-hand-created-manifest warning with their reason. Cargo generates the expected workspace-inheritance entries. Explaining the mechanism makes the existing requirement easier to preserve without a second prohibition.
.agents/skills/public-dependency/SKILL.md Remove generic cargo-add syntax; lead directly with the # public marker. The non-obvious repository rule is the public-API annotation and its propagation consequences, not how to invoke Cargo. The marker, exposure cases, and dependency policy remain.
.agents/skills/protocol-reviewer/SKILL.md Condense repeated review framing, use one sentence per line, and add the moved STYLE pointer. The skill retains authoritative-source fallback, requirement mapping, normative/informative distinction, and exact citations; the PDU-specific pointer is now loaded only for the relevant review.
.github/skills/code-review/SKILL.md Add explicit root/ancestor AGENTS and applicable intent reads before selecting passes. A review can miss nested policy if it selects its checks from the root alone. This makes instruction discovery an explicit prerequisite.
.github/skills/release-prep/SKILL.md Replace hardcoded origin/main with the PR's actual base; consolidate public-dependency and changelog procedures. The repository default is master, so the old base was wrong. Delegating the marker definition prevents drift while retaining bump propagation, net-release changelog rules, exact generated formatting, narrow edit scope, and lockfile checks; generated links are correctly described as inline links.
.agents/skills Refresh seven configured skill directories from Gateway. The changed mirrors are commit-and-pr-conventions, contract-review, intent-drift-audit, intent-files, intent-review, test-oracle-reviewer, and test-pruner (including references). They are copied unchanged to restore upstream parity, not locally re-audited or customized.
.claude/skills Replace the whole-directory symlink with 42 individual skill symlinks. The whole-folder link is not discovered by Claude. Per-skill links expose shared skills plus local code-review and release-prep without duplicating their contents.

Human-owned intent files and application code are unchanged.

Shared heading and prevention-rule changes originate in Gateway #2163.
Merge that source PR first so scheduled skill synchronization preserves the cleanup.

Note

Human-tuned, LLM-assisted content.

@github-actions github-actions Bot added scope/web Affects the web/WASM ecosystem and removed needs-author-action The pull request author is the current next actor labels Oct 27, 2026
Keep repository guidance accurate and concise, refresh shared skills
from Gateway, and expose every skill through an individual Claude symlink.
Remove redundant skill-discovery and synchronization boilerplate from
the root instructions.
@github-actions github-actions Bot added the needs-author-action The pull request author is the current next actor label Oct 27, 2026
Remove opening banners that repeat skill metadata or repository context.
Synchronize the shared authoring and review rules that prevent redundant
titles while retaining headings that organize instructions or add context.
@CBenoit
Benoît Cortier (CBenoit) force-pushed the codex/agent-instructions-maintenance branch from a354b2b to 1dc1df5 Compare October 27, 2026 12:28
@github-actions github-actions Bot removed the needs-author-action The pull request author is the current next actor label Oct 27, 2026
Keep test-oracle review and pruning rules aligned with the shared
Gateway definitions.
@CBenoit
Benoît Cortier (CBenoit) merged commit d78599f into master Oct 27, 2026
44 checks passed
@CBenoit
Benoît Cortier (CBenoit) deleted the codex/agent-instructions-maintenance branch October 27, 2026 15:20

This branch was previously deployed

1 inactive deployment
llm-providers — f65ac08f Deployed Oct 27, 2026 by CBenoit via Classify pull request #9325
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

automation-failed Exact-head automated classification or review failed or was unavailable risk/unknown Risk could not be determined automatically; needs maintainer-level scrutiny scope/tooling Build, CI, release, or developer tooling scope/web Affects the web/WASM ecosystem size/XXL Size: 1300 or more counted lines or 50 or more files

Development

Successfully merging this pull request may close these issues.

1 participant