Drop the unconditional gh repo view lookup. Step 1 already reads the
log, which shows whether the repo uses merge commits or a linear
history, so the strategy comes free; the lookup is now the fallback
for a rejected merge. Also record why an explicit strategy is required
at all: gh errors out when nothing can prompt for one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
User-invoked engineering skill that takes the working tree from
uncommitted changes to a merged PR: batch commits by intent, branch
off the default branch, open the PR, merge with an allowed strategy,
then prune the branch locally and on the remote.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Center the header with tagline and badges, rename title to simonwong/skills
- Put Bun and npm install commands side by side
- Collapse the motivation section and the writing-skill references
- Mark invocation mode with icons in the skill tables
- Drop the horizontal rule
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
write-article covers the collaboration flow from idea to finished draft:
material inventory with a supplement loop, direction confirmation, then
drafting in a fixed order (body, opening, ending, title, layout). Review
and AI-flavor check run in fresh subagents.
writing-ai-check diagnoses expression in a single draft. Its input and
its evidence are the draft itself — it never judges authorship, never
outputs an AI rate, and never verifies external facts.
Each reference file answers one kind of question; feature lists live only
under writing-ai-check/references/. Decisions recorded in docs/adr/.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Record the issue tracker (GitHub Issues via gh) and the single-context
domain doc layout under docs/agents/, and point to them from AGENTS.md.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- Clarified the structure and organization of skills in AGENTS.md, including the addition of in-progress skills.
- Revised README.md to reflect changes in writing skills, emphasizing the workflow and specific skill functionalities.
- Introduced a new section for in-progress skills, detailing their purpose and data contracts.
- Enforce read-before-write: must read index.jsonl + ls entries/ before writing
- ID numbering now checks both index tail and entries/ directory, takes max+1
- Explicitly forbid overwriting index.jsonl with Create/Write tools
- Bump both skills to 1.2.0
JSONL (one JSON object per line) is simpler for append operations -
no need to read the whole array, parse, append, and write back.
Just append a new line. Updated all 4 skills that reference index files:
material-ingest, material-retrieve, compose, style-extract.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add explicit JSON special character escaping notes to material-ingest and
style-extract to prevent broken JSON when content contains double quotes
- Remove redundant index.md from material-ingest (index.json is sufficient)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Simon