Install
$ agentstack add skill-tkhribech-claude-founder-os-session-close ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Session Close — write the baton
The read-back half of this is the session-start block in START-HERE.md; this skill is the write half. Its one job: make the next session start from knowledge, not from zero. Run the steps in order.
Hard rules (apply to every step)
- Memory-only. No live-system changes at close — read, capture, organize, write files. Nothing outward-facing.
- Capture durable signal only. A one-off ask (a quick draft, a single lookup) stays chat-only — no memory file for it or for the people/tools it mentioned once. Persist only reusable knowledge and things the founder will reference again. Unsure → skip; over-capturing creates orphans.
- Delete, don't archive. Superseded or empty files go to the system trash — no
-oldcopies, no archive folders in the workspace; stale archives are how dead rules get trusted as current. Merge nuance into the survivor first. - Verify, don't assert. Any "clean/confirmed" claim in the close summary must be backed by a command actually run this session.
- Date-stamp everything touched (first line = today, absolute date).
Steps
- Inventory today's touches. List the files created or edited this session (from the session itself, not just filesystem timestamps).
- Capture the signal — THE anti-repeat step. Extract from this conversation: (a) every correction the founder made or better-way that got confirmed → a dated file in
memory/feedback/(body: the incident, Why:, How to apply:); (b) every durable decision or ruling → update the matching file inmemory/RULES.md/memory/project/, dated. Search for an existing file first — update it rather than spawning a near-duplicate. - Consolidate duplicates. If today produced two files covering one fact, merge into the canonical one (full-read both first), fix inbound links, trash the loser.
- Index for next session. Add one-line pointers for today's new items to
memory/MEMORY.md(rolling headline + the relevant anchor), and bump its date. If the index is over its size cap, compress the oldest entries into their linked files — the index stays pointers-only. - Update the open loops.
memory/status/PENDING-TASKS.md: add new items, mark done ones, re-date deferred ones. - Write the handoff.
memory/status/session-handoff-.mdper the template: what shipped (verified, with evidence), what's in flight with its exact next step, decisions made, landmines, open questions for the founder. Written for a reader with zero context — include full paths and absolute dates. - Hygiene sweep. Trash today's scratch/zero-byte/untitled files — never files another session may own (when in doubt, leave it); date-stamp anything missing it.
- Verified close summary. Print: files touched/created/merged/trashed · corrections + decisions captured (by name) · index updates · warnings (conflicts, skips, failures) — each "clean" claim paired with the command that proved it.
The mid-session habit this skill assumes
Don't save all capture for close: after every completed task, update the affected tracker or project file while context is fresh. Close then becomes a sweep, not an excavation — if closing regularly takes more than a few minutes, capture is happening too late.
Failure mode to avoid
Fail soft: if any step errors, log the warning and continue — never let one failed command halt the close. An imperfect handoff beats no handoff; the next session's cold start is the one reader you must not leave empty-handed.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: tkhribech
- Source: tkhribech/claude-founder-os
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.