Claude MD Updater
Turns session findings into a reviewable CLAUDE.md diff, filtering out transient notes
Test report
- Verdict
- Tested · Works
- Score
- Tested
- Jul 31, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · 1fe635c
Cloned the repo and installed into a throwaway HOME; SKILL.md landed at ~/.claude/skills/claude-md-updater/SKILL.md, frontmatter has name+description, the body is 97 lines with zero script or file references and no curl|sh, base64, or injection text. Output test: I wrote a fixture CLAUDE.md for a Rails "ledger service" plus a condensed session transcript containing four items (an OOM root cause, a new preview environment whose transcript line included a literal DB password, a "I run Ruby 3.2 not 3.3" personal note, and a "this test was flaky but passed on 3 reruns" note), then produced a baseline update and a skill-followed update. Baseline rewrote the whole file, invented a "## Notes" section holding both the Ruby-version and flaky-test lines, and carried the literal secret through as 'entry "ledger-preview-db" (hunter2-preview-9981)'; the skill version emitted a unified diff touching only three sections, ended with "Apply this diff? Nothing is written until you approve.", wrote the credential as 'entry "ledger-preview-db"' with no value, and explicitly listed the two transient items as excluded with a destination for each. Caveat I'll state plainly: I authored both artifacts, so the baseline is my honest best guess at unguided behavior rather than a blind run — the excluded-transient-notes and diff-vs-full-rewrite deltas are solid, the password carry-through is plausible but arguable. Trigger phrasings judged: SHOULD fire — "Update CLAUDE.md with what we learned in this session" (yes), "Save this lesson so future sessions don't repeat it, add it to our project context file" (yes), "What should we remember from today's debugging? Persist it" (yes); SHOULD NOT — "Read CLAUDE.md and tell me what build command this project uses" (correctly no, read-only), "Summarize what we did this session for my standup notes" (correctly no, no persistence target). 5/5.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 7/10
- Docs & honesty 4/5
What Claude MD Updater does
Scans the current conversation for durable facts — root causes found while debugging, new endpoints and paths, recurring commands — and proposes them as a unified diff against the project's CLAUDE.md, writing nothing until you approve. Triggers on "update claude.md", "save this lesson", "what should we remember", or "add to context". Its rules also keep session-specific notes, personal environment quirks, and credential values out of the git-committed file, routing them to auto memory or CLAUDE.local.md.
How to install Claude MD Updater
git clone --depth 1 https://github.com/jamditis/claude-skills-journalism.git /tmp/claude-md-updater-src
mkdir -p ~/.claude/skills
cp -R /tmp/claude-md-updater-src/dev-toolkit/skills/claude-md-updater ~/.claude/skills/claude-md-updater
# Single SKILL.md, no scripts, no dependencies, no API keys.
# Plugin alternative: the repo ships a marketplace (.claude-plugin/marketplace.json);
# /plugin marketplace add jamditis/claude-skills-journalism then install the "dev-toolkit" plugin,
# which bundles this skill alongside 6 others.
# The body cross-references a "project-memory" skill that lives in a DIFFERENT toolkit in the
# same repo (project-templates-toolkit/skills/project-memory) — not installed by the block above.
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Claude MD Updater
-
/claude-md-updaterTurns session findings into a reviewable CLAUDE.md diff, filtering out transient notes
It also activates on plain-language prompts like these:
-
Update CLAUDE.md with what we learned today -
Save this lesson before we end the session -
What should we remember from this conversation
Frequently asked questions
- Is the Claude MD Updater skill free?
- Yes. The skill itself is free from jamditis/claude-skills-journalism. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Claude MD Updater work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 31, 2026. Verdict: Tested · Works. Cloned the repo and installed into a throwaway HOME; SKILL.md landed at ~/.claude/skills/claude-md-updater/SKILL.md, frontmatter has name+description, the body is 97 lines with zero script or file references and no curl|sh, base64, or injection text. Output test: I wrote a fixture CLAUDE.md for a Rails "ledger service" plus a condensed session transcript containing four items (an OOM root cause, a new preview environment whose transcript line included a literal DB password, a "I run Ruby 3.2 not 3.3" personal note, and a "this test was flaky but passed on 3 reruns" note), then produced a baseline update and a skill-followed update. Baseline rewrote the whole file, invented a "## Notes" section holding both the Ruby-version and flaky-test lines, and carried the literal secret through as 'entry "ledger-preview-db" (hunter2-preview-9981)'; the skill version emitted a unified diff touching only three sections, ended with "Apply this diff? Nothing is written until you approve.", wrote the credential as 'entry "ledger-preview-db"' with no value, and explicitly listed the two transient items as excluded with a destination for each. Caveat I'll state plainly: I authored both artifacts, so the baseline is my honest best guess at unguided behavior rather than a blind run — the excluded-transient-notes and diff-vs-full-rewrite deltas are solid, the password carry-through is plausible but arguable. Trigger phrasings judged: SHOULD fire — "Update CLAUDE.md with what we learned in this session" (yes), "Save this lesson so future sessions don't repeat it, add it to our project context file" (yes), "What should we remember from today's debugging? Persist it" (yes); SHOULD NOT — "Read CLAUDE.md and tell me what build command this project uses" (correctly no, read-only), "Summarize what we did this session for my standup notes" (correctly no, no persistence target). 5/5.
- What is the Claude MD Updater SkillProof Score?
- 8.4/10 — installs cleanly 5/5, triggers reliably 5/5, output vs. baseline 7/10, docs & honesty 4/5.
- How do I install Claude MD Updater?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Claude MD Updater with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.