recall · pwguler — Claude Code skill: project context and installation
recall — a Claude Code skill that rebuilds recent project context from agent sessions, the live repository, and project history. Run it before starting or resuming work, with a topic or without one.
MIT 922 installs ★ 10 updated 2026-10-16
npx skills add pwguler/skills --skill recall -a claude-codeWhat recall does
recall gathers the current state of a project from earlier agent sessions, the working repository, and its history, then returns a short brief. It searches session transcripts for the requested topic and checks discovered branches, pull requests, commits, and tickets against the current state with git and gh.
The brief covers current progress, open threads, recurring problems, and one next action. For a feature, file, code area, or bug, the skill also uses the history workflow from research. It only reads: files, branches, and tickets remain unchanged.
What it helps with
- recover recent project context from agent sessions
- check discovered pull requests and commits against the live repository
- collect open threads and recurring problems in a short brief
- identify a concrete next action from a task's history
Example prompts for the agent
- Recall where work on this task stopped and list the open threads.
- Summarize work on
src/auth.tsfrom earlier sessions and repository history. - Before continuing, check the discovered branches and pull requests against the current project state.
How to install recall
1Skills CLI (Vercel)
npx skills add pwguler/skills --skill recall -a claude-code2Manually: put the files in place
~/.claude/skills/recall/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/recall/
1Skills CLI (Vercel)
npx skills add pwguler/skills --skill recall -a codex2Manually: put the files in place
.agents/skills/recall/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add pwguler/skills --skill recall -a cursor2Manually: put the files in place
.cursor/skills/recall/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install pwguler/skills skills/recall2Manually: put the files in place
.github/skills/recall/SKILL.md1Skills CLI (Vercel)
npx skills add pwguler/skills --skill recall -a gemini-cli2Platform CLI
gemini skills install https://github.com/pwguler/skills --path skills/recallManually: put the files in place
.devin/skills/recall/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add pwguler/skills --skill recall -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about recall
How do I install recall in Claude Code?
Run npx skills add pwguler/skills --skill recall -a claude-code. The skill file is at ~/.claude/skills/recall/SKILL.md.
What does recall do?
It builds a short project status brief using earlier agent sessions, the working repository, and project history. The brief covers current progress, open threads, recurring problems, and a next action.
Does recall change files or branches?
No. The skill reads project data but does not modify files, branches, or tickets.
Source text
SKILL.md · MIT
--- name: recall description: Rebuild where recent work stands, from past sessions of the agent, the live repository, and the history of the project. Give back a short brief. Run it before you start or continue work, with or without a topic. disable-model-invocation: true --- Before work starts or continues, rebuild where it stands, and give back a short brief. Read only what the topic needs, then stop. Fast path: when the message of the user already holds the state, use it and skip the session search. The state is paths, the branch, and the change in progress. 1. Fix the scope, and say it back in one line. The window is the last 7 days, unless the user names another. Add the topic when one is named. The project is this working directory. Read the sessions of another project only when the user asks. "All" stays all. Never reduce it to the latest few without saying so. 2. Find the transcripts of the past sessions of this project, as [transcripts.md](references/transcripts.md) says for your agent. Sort them by modification time (`ls -t`). Leave out the current session and the transcripts of subagent runs. When your agent keeps no transcripts that you can read, say so, and work from the repository and its history. 3. Search before you read. Grep each transcript for the topic, then read only the lines that match. Read the summaries that the agent already wrote before the turns. When the matching lines total more than about 2,000, split the transcripts across subagents. Each subagent returns one block for each session. The block gives the topic, the user's goal, decisions, open threads, the user's corrections, and artifacts. Artifacts are branches, PRs, commits, and tickets. Tag each block with its session id. 4. When the topic names a feature, file, area, or bug, also run the history path of the `research` skill. Ask what the current state is, what was tried and did not work, and what users still report. 5. Check every branch, PR, commit, and ticket that the search found against the live state, with `git` and `gh`. Read what an agent did from its session entries: the commands that it ran, and the errors that it hit. Never infer it. 6. Write the brief in four sections. **Where it stands** has five bullets at most. **Threads** has one line for each thread. Each line starts with exactly one tag: `[in flight <branch>]`, `[open PR #N]`, `[merged #N]`, `[reverted #N]`, `[done, uncommitted]`, or `[planned]`. **Problems** has five items at most: the problems that recur. Include complaints that users still make, and any fix that landed and was reverted later. **Next move** is one concrete action. Rules: - Stay on the topic. A thread next to it goes in only when it blocks this one. When the brief is longer than one screen, cut detail before you cut threads. - Cite every finding: a session by its id, and the history by its source (commit, PR, or ticket). - Recall only reads. It changes no file, branch, or ticket. - Redact secrets found in a session. Replace a token, key, password, or connection string with a placeholder.
© pwguler · MIT · pwguler/skills
Similar skills
SKILL.md skills catalog →recall by other authors comes first.