agentleFS
IlyaGorskyIlyaGorsky/memory-toolkit/skills/session-start/SKILL.md

session-start

Skill13 starsChanged 11 days agoMITView on GitHub

Cold session start — context, status, focus. Use at the beginning of each new session.

39 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use deploy-verify, autopilot and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • danielarif26Skill

    deploy-verify

    Verify a Vercel deployment actually went live and serves the correct production domain. Self-improving: reads its own run history before checking and logs the outcome after. Use when the user asks "did you deploy?", "is it live?", "verify the deploy", "check vercel", or after any Vercel deploy. Part of the Agentic OS (Level 1 backbone).

    Also in 3 projects that use this

  • ibahgatSkill

    autopilot

    Run autonomous multi-stage execution loops until acceptance criteria pass or blockers are reached.

    Also in 3 projects that use this

  • MicrosoftSkill

    handoff

    Compact the current conversation into a handoff document for another agent to pick up.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 32 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a workstream argument is provided — skip to Step 3.

    line 20
  2. Very specific:

    Run each command as its own Bash call — that is what the skill's allowed-tools pre-approves (no variables, loops or pipes):

    line 22
  3. Distinctive:

    If user picks ➕ — ask name + keywords, then:

    line 63
  4. Very specific:

    If decisions on this theme are already in the notes (node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" unassigned), offer a card drafted from …

    line 67
  5. Very specific:

    No match → run git log <branch> --oneline -10 + git diff $(git merge-base HEAD "origin/$(git symbolic-ref --short refs/remotes/origin/HEAD …

    line 84
  6. Very specific:

    If handoff has lastCommit ref: show git log <ref>..HEAD --oneline (changes since session-end).

    line 86
  7. Distinctive:

    Read the card first: it frames the handoff.

    line 99

“Common” is measured: kept by at least 6 in 10 of the 32 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/session-start — Cold session start

Quick context entry. Gather state, propose focus.


Step 1: Gather context

Do NOT re-read MEMORY.md — already in system prompt via claudeMd loader.

If a workstream argument is provided — skip to Step 3.

Run each command as its own Bash call — that is what the skill's allowed-tools pre-approves (no variables, loops or pipes):

node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" handoff        # latest handoff (per-workstream ones in Step 3)
node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" workstreams    # sorted by aggregate weight
node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" dir            # memory folder, keyed by git root like CC

Then read the newest notes/YYYY-MM-DD.md in that folder (first ~30 lines), and for each of the top 3 workstreams:

node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" workstream <name> --brief

Decision tree — source of focus candidates:

Handoff stateCandidates source
≤7 days old, has "What's next"Handoff primary (workstreams table still rendered for context)
>7 days oldWorkstreams table; handoff shown as stale context
Missing/empty/only SESSION_START markersWorkstreams table; say "no usable context", suggest /session-restore
No handoff, no notesWorkstreams table only

The workstreams table in Step 1b is always rendered in Step 4 — it's the session's map, not a fallback. The decision tree only controls where focus candidates come from.


Step 1b: Workstreams table (always rendered)

Render as table. Top-3 rows show synthesized "What's in" from --brief; rest show keywords. ➕ row is mandatory.

**Workstreams** (sorted by aggregate weight)

| # | Workstream | Last touched | What's in / Keywords |
|---|---|---|---|
| 1 | <name> | <date> | <synthesized 1-line for top-3> |
| ... | | | <keywords for rest> |
| N | ➕ Create new | | |

If user picks ➕ — ask name + keywords, then:

node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" add-workstream <name> <keyword1> <keyword2> ...

If decisions on this theme are already in the notes (node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" unassigned), offer a card drafted from them — see card in the memory skill.


Step 2: Git status

git status; git log --oneline -5; git branch --show-current
  • Uncommitted changes → ask: "Commit or continue on top?"
  • Last commit >5 days ago → note it
  • Commits: synthesize last 2-3 into 1 line about the theme. Don't list verbatim.

Branch ↔ workstream check: does branch name contain any keyword from workstreams.json?

  • Match → skip "Current branch" block in Step 4
  • No match → run git log <branch> --oneline -10 + git diff $(git merge-base HEAD "origin/$(git symbolic-ref --short refs/remotes/origin/HEAD | sed 's|^origin/||')")...HEAD --stat (default branch, not hardcoded main). Synthesize phase (review iteration / new feature / hotfix). Grep notes/ for branch name. Surface in Step 4 as branch block + extra focus candidate.

If handoff has lastCommit ref: show git log <ref>..HEAD --oneline (changes since session-end).


Step 3: Workstream brief

node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" card <name>         # what the workstream is, what's open — NO_CARD if none
node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" parked <name>       # what was set aside in it — mention titles in the task plan
node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" handoff <name>      # per-workstream handoff if available
node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" workstream <name> --brief

Read the card first: it frames the handoff. Show: open/done task counts, blockers, pickable now. Do NOT read full files — full context loads only after task pick in Step 4.


Step 4: Render & propose

Use the user's preferred language (from profile/language.md if present, else English). Translate labels and prose; keep markdown structure, identifiers, code refs as-is.

Required format — clean bold headers, no emoji, --- between every block, blank line after each header before content.

## Session YYYY-MM-DD

**Git:** branch `<branch>` / <N> uncommitted / last commit <days> days ago
**Commits:** <1-line synthesis>

---

**Workstreams**

[table from Step 1b]

---

**Task plan ({active workstream})**

{1-line context}
- [ ] {open task} ({SP})
- [ ] {open task} — {blocker}

---

**Current branch `{branch}`** _(only if no workstream match)_

{1-2 sentences: theme, phase, side task vs portfolio}

---

**Focus candidates**

1. **{task}** — {why now} → `/model {model}` ({reason})
2. **{task}** — {why now} → `/model {model}` ({reason})
3. **{task}** — {why now} → `/model {model}` ({reason})

---

My hypothesis: focus on **{candidate #1}** because {reasoning grounded in handoff/git/recency}. Recommendation: `/model {model}`.

Agree, pick another candidate, or **start a new workstream**?

Close with a hypothesis, not a menu. Users iterate on a proposal faster than they pick from a list. Candidates 2-3 stay visible for quick override.

MANDATORY: the closing line must keep "start a new workstream" as an explicit option. Never omit it — the hypothesis frame hides existing workstreams, so the escape hatch to create a new one has to be surfaced every time. If the user picks it, run Step 1b (workstream creation).

Model selection: Opus = planning/architecture/exploration. Sonnet = execution by plan, refactoring, review. Haiku = routine/translation/pattern copy. Write it as the command to type — /model opus, /model sonnet, /model haiku — one action instead of a menu.


Step 5: Session label

node "${CLAUDE_PLUGIN_ROOT}/scripts/memory.js" note "SESSION_START branch:$(git branch --show-current) workstream:<chosen> focus:<task>"

(Auto-routed to sessions.log, not notes.)


Rules

  • Max 3 focus candidates
  • Skip blocked tasks (waiting on mockups/backend) from candidates — note them separately
  • Workstream argument → go straight to Step 3
  • Previous session ended with compact → check via /session-restore

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.