Clarify ambiguous pronouns in communication guidelines - #6
Merged
Conversation
Replace "I/me/my" (meaning the user) with explicit "the user" referents so Claude doesn't misread who has decision authority — especially in "Disagree but commit" where "I make the final call" was ambiguous. https://claude.ai/code/session_01HB5kAroaJwVvju6vATCfUq
Warmer language that's consistent with the "partners solving problems together" framing already established in the file. https://claude.ai/code/session_01HB5kAroaJwVvju6vATCfUq
5 tasks
ahoym
added a commit
that referenced
this pull request
Mar 14, 2026
…l rule - Remove 3 generic OOP sections from code-quality-instincts.md (no session evidence they were missed) and fix missing newline at EOF - Remove rm -rf staging cleanup from extract-request-learnings (let next run's mkdir -p handle it) - Add glab --body-file limitation note to gitlab-commands.md - Add "never dismiss comments as duplicates based on topic" guidance to address-request-comments skill Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ahoym
added a commit
that referenced
this pull request
Mar 14, 2026
* Add quiet no-op for polling, LGTM post-verification nudge, and skill design learnings - Quiet no-op: incremental fetches with 0 new comments emit a single line instead of full formatted output (saves context in polling loops) - After LGTM verification: note that further monitoring is unlikely needed - New learnings: @ reference format flexibility, body-only templates, skills shouldn't assume invocation context Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Curate skill-design.md: prune, merge, and migrate patterns Delete 5 patterns covered elsewhere (address-request-comments, curation-insights, content-type-decisions). Merge 6 related patterns into 2 (@ references, producer-consumer). Migrate 2 patterns to better homes (multi-agent-patterns, cross-repo-sync). Clean up stale git:create-mr references. Net: 38→31 patterns, ~14% line reduction. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add @ ref format findings, migrate reply files to .gh-replies/ directory Empirically verified that @ references in SKILL.md resolve regardless of formatting (list prefix, description text). Migrated reply file pattern from single .gh-reply.tmp to .gh-replies/<id>.md directory for parallel safety and queryability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Update learnings, guidelines, and curate skill from recent sessions Capture patterns from recent session retros: git workflow gotchas (pagination, symlinks, stacked PRs), Python practices (dataclasses, pyproject.toml, noqa), PR review etiquette, guideline taxonomy, and curate skill decomposition format for large files. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add staging directory pattern, curation refinements, and session learnings Extract-request-learnings: use project-local staging dirs for general/private writers to work around background agent write restrictions on ~/.claude/. Orchestrator copies to final locations. Curate classification: refine "migrated completely" to check whether the destination is broadly discoverable before deleting source. New guideline: don't ask permission to invoke sub-skills within a skill's instructions. New learnings on partial batch completion and scope classification language-awareness. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add git sync step, language routing, content taxonomy, and session learnings Extract-request-learnings: add git fetch step for multi-session staleness, language tag to extractor output, route-check step to writer for language-specific file routing. New guideline: surface known limitations before acting, not after. New learning: content type taxonomy for Claude config surface. New patterns: staging directory for out-of-project writes, multi-session skills need remote sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add confidence-level gate and skill design learnings New soft gate in context-aware-learnings: self-check whether working from loaded knowledge or training memory before drafting substantial domain content. Experimental — relies on self-awareness, not mechanical triggers. Skill design learnings: hub-and-spoke for authoring guides, reviewer personas as distinct from working-in-repo personas, memory minimalism principle. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add learnings lookup step to address-request-comments skill New step 6 globs learnings/ filenames and reads domain-relevant files before composing replies, grounding responses in established knowledge rather than ad-hoc reasoning. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Update learnings, skills, guidelines, and references from recent sessions - Refine path resolution table in CLAUDE.md for clarity - Update extract-request-learnings, create-pr, create-request, and resolve-conflicts skills - Expand communication guidelines - Update learnings across ci-cd, claude-authoring, code-quality, git, process, and testing domains - Update GitHub/GitLab skill reference docs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add surgical conflict reading and template re-read learnings - resolve-conflicts: grep for conflict markers instead of full file reads, use offset+limit to read only conflict regions - claude-authoring-skills: re-read templates at point of use, not ahead of time, to prevent stale-context improvisation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Split curate skill into mode-specific reference files Extract content-mode and skill-mode execution steps, report templates, and apply actions into conditional reference files. SKILL.md drops from 565 to 186 lines (always loaded), with each mode file loaded only when relevant. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Generalize .gh-replies permission to cover all repos under ~ Broadens the Write permission pattern from ~/WORKSPACE/**/.gh-replies/** to ~/**/.gh-replies/** so repos outside WORKSPACE are also covered. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Rename .gh-replies to change-request-replies for platform agnosticism The directory name was GitHub-specific but the skills support both GitHub and GitLab. Also switches gh api URL templates from double quotes to single quotes to avoid zsh permission prompts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Drop URL query params from incremental fetch templates Move timestamp filtering to jq client-side instead of using ?since= and &direction= query params. URL query params require quoting which triggers permission prompts in Claude Code. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Move broad sweep flow to content-mode.md and add decomposition learnings - Relocate broad sweep execution steps from SKILL.md to content-mode.md where they belong as a content mode variant - Add skill decomposition learnings: execution path splitting, mode variants stay with parent mode, single-file anti-pattern Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add note to avoid bare #N in review replies GitHub auto-links #1, #2, etc. to issues/PRs. Use backtick-wrapped or plain numbering instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Remove duplicated sections and fix copy-paste takeaway - Remove spoke-duplicated sections from claude-authoring-guidelines.md (Avoid Nesting, Content Type Taxonomy, Genericize, Persona-Learning Boundary Test) — already in their authoritative spoke files - Remove duplicated "behavioral rules" paragraph from communication.md — already in CLAUDE.md MEMORY section - Remove duplicated scope classification section from claude-authoring-skills.md — belongs in claude-authoring-learnings.md - Fix copy-paste takeaway under Memory Minimalism that belonged to Multi-Session Skills section Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add code-review-request skill and inline review API references - New git:code-review-request skill for reviewing PRs/MRs - Add inline comment review API patterns to GitHub and GitLab command references Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add review count tracking and emoji reactions to address-request-comments - Track LAST_REVIEW_COUNT for incremental review submission detection (reviews endpoint doesn't support since/updated_after filtering) - Include review count in quiet no-op check - Add emoji reactions for positive signals/general feedback instead of text replies - Add React to Comment templates to both github and gitlab commands Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add --paginate to all gh/glab api fetch templates Dropping query params (per_page, direction) to avoid quoting caused comments beyond the default 30-per-page limit to be silently missed. --paginate is a CLI flag (no quoting needed) that auto-fetches all pages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Fix React to Comment template: use -f not -F for emoji reactions -F infers type and treats +1/-1 as numeric, which the GitHub API rejects. -f (lowercase) always sends as string. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Implement review suggestions #6, #7, #8 and add no-duplicate-dismissal rule - Remove 3 generic OOP sections from code-quality-instincts.md (no session evidence they were missed) and fix missing newline at EOF - Remove rm -rf staging cleanup from extract-request-learnings (let next run's mkdir -p handle it) - Add glab --body-file limitation note to gitlab-commands.md - Add "never dismiss comments as duplicates based on topic" guidance to address-request-comments skill Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add structured footnote with Role field for self-reply filtering Replace generic "Co-authored with Claude" footer with structured metadata: model, persona, and role. The Role field (Addresser for this skill, Reviewer for code-review-request) enables filtering own replies by role instead of fragile content matching. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Add re-review mode and structured footnote to code-review-request skill - Auto-detect previous reviews by matching Persona + Role in comment bodies - Re-review evaluates author responses: resolved (emoji react), partially addressed (follow-up reply), not addressed (re-raise) - Reviews new commits since last review as delta analysis - Standardize footnote format: Co-Authored, Persona, Role fields - Filter on Persona + Role composite key to separate reviewer vs author comment chains from the same persona - Summary = themes only, inline comments = specifics (no duplication) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Add gh api query param workaround and GitHub reviews filter learning - bash-patterns: gh api URLs with query params trigger permission prompts; use client-side jq filtering instead - git-patterns: GitHub reviews endpoint has no since filter, track review count for incremental fetches Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Restore OOP patterns to code-quality-instincts.md These were scraped from actual agent implementation sessions and have session evidence. Premature removal reversed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Refine code-review-request skill and add review workflow learnings - code-review-request: add quick-exit check for polling, reference platform commands by section name instead of inlining, add notes on empty reviews and footnote identity - bash-patterns: gh api pagination, --input for JSON, -f vs -F - claude-authoring-skills: structured footnotes, re-review detection, polling fresh-fetch rule, quick-exit pattern, platform command refs - claude-code: GitHub reviews API single-payload pattern - process-conventions: multi-agent comment identity, review summary vs inline separation, emoji reactions, empty review handling Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Restore missing learnings sections across 4 files Add back gaps to claude-authoring-skills.md (user interaction points, file operations, portability, token optimization, half-steps, validation checklist, bash commands, description frontmatter, routing hints). Add diagnostic sequence and skill prerequisites to claude-code.md. Add orchestrator/agent split and assumption verification to multi-agent-patterns.md. Add on-platform approval flows to process-conventions.md. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Add merged/closed state check to code-review-request skill Skip review and suggest cancelling cron job when PR is merged or closed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
ahoym
added a commit
that referenced
this pull request
May 17, 2026
Addresses team review (#1-#6) on PR #109. - Add `[lib]` target. `Provider`, `home`, `expand_tilde`, and `providers(include_project_local: bool)` move to `src/lib.rs`. `SCHEMA_VERSION` becomes a single `u64` constant. Both binaries import via `learnings_suggest::*`. - `index-build.rs` calls `providers(false)` with a comment documenting the intentional projectLocal omission (CWD-dependent, runtime-only). The behavioral split is now explicit at the call site. - `anchor_commit(provs: &[Provider])` takes the already-loaded slice instead of re-parsing learnings-providers.json. - Decompose `run()` (~200 lines) into `match_sections`, `match_files`, `merge_and_dedup`, `compute_forced`, `format_block`. `run()` is now a ~35-line coordinator; each phase has a typed input/output and is testable in isolation. - `extract_quoted`: comment notes ASCII-only matching (macOS smart-quote autocorrect is not captured). - Linux x86_64 binaries rebuilt via `cargo zigbuild`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ahoym
added a commit
that referenced
this pull request
May 18, 2026
Addresses team review (#1-#6) on PR #109. - Add `[lib]` target. `Provider`, `home`, `expand_tilde`, and `providers(include_project_local: bool)` move to `src/lib.rs`. `SCHEMA_VERSION` becomes a single `u64` constant. Both binaries import via `learnings_suggest::*`. - `index-build.rs` calls `providers(false)` with a comment documenting the intentional projectLocal omission (CWD-dependent, runtime-only). The behavioral split is now explicit at the call site. - `anchor_commit(provs: &[Provider])` takes the already-loaded slice instead of re-parsing learnings-providers.json. - Decompose `run()` (~200 lines) into `match_sections`, `match_files`, `merge_and_dedup`, `compute_forced`, `format_block`. `run()` is now a ~35-line coordinator; each phase has a typed input/output and is testable in isolation. - `extract_quoted`: comment notes ASCII-only matching (macOS smart-quote autocorrect is not captured). - Linux x86_64 binaries rebuilt via `cargo zigbuild`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Replace "I/me/my" (meaning the user) with explicit "the user" referents
so Claude doesn't misread who has decision authority — especially in
"Disagree but commit" where "I make the final call" was ambiguous.
https://claude.ai/code/session_01HB5kAroaJwVvju6vATCfUq