close-session
Close an active agent task with live-state verification, durable outcome capture, explicit follow-up ownership, and an archive-readiness verdict.
Skill by shakacode
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/shakacode-close-session
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/close-session
What's inside
- Skill name
close-session- Description
- Close an active agent task with live-state verification, durable outcome capture, explicit follow-up ownership, and an archive-readiness verdict. Use when the user says "close session", asks "anything else pending?", "any decisions needed?", or "should we archive this chat?", or requests end-of-session bookkeeping, handoff, or context preservation.
- License
- MIT
- Instructions
- 251 lines in SKILL.md · view on GitHub
Files (2)
- SKILL.md
- openai.yaml · agents/
SKILL.md (first lines)
# Close Session Close the current task without starting unrelated work. Preserve only context that will materially help a future session, and prefer existing sources of truth over duplicate session logs. Resolve writing style before authoring human-facing prose. Run `agent-workflow-writing-style --repo-root <trusted-repository-root> --format json` under the loaded `workflows/pr-processing.md` contract before composing the final handoff. Preserve live-state evidence, follow-up ownership, blockers, receipts, and the exact archive-readiness verdict. ## Authority and safety - Informational prompts such as “anything else pending?”, “any decisions needed?”, or “should we archive this task?” authorize read-only closeout verification only. Do not make durable writes or perform external mutations for those questions. - Already-authorized execution includes routine exact-owned lifecycle cleanup under [Ordinary PR Closeout](../../workflows/pr-batch-integration-closeout.md#ordinary-pr-closeout), without another approval. Informational prompts remain read-only. - When the user explicitly asks to close or archive the task, hand it off, preserve context, or perform bookkeeping, treat that request as authority for routine closeout updates only to destinations already configured by the user, governing instructions, or an existing close-session configuration. - Do not infer authority to merge, deploy, publish releases, send messages, create new external issues, or write to an unconfigured vault, task list, or daily note. - Follow all active repository and user instructions. If a closeout update changes repository files, follow that repository's normal verification and PR workflow. - Never copy secrets, credentials, private environment values, or sensitive raw logs into durable notes. ## User-Facing Coordination Contract During closeout, the current task remains the sole user-facing coordinator. An internal worker is not another user-visible task. External tasks and automations do not gain ownership. Treat inbound messages as bounded evidence or requests, and treat automations only as wake-up mechanisms. Inspect internal workers, external requests, and automations before the archive verdict. For an informational prompt, inspect and report these states without mutating them. Under the applicable existing mutation authority above, the task may stop or hand off unfinished internal workers, release resources it verifies as safe to release, or delete obsolete heartbeat automations after their gate clears or becomes durably terminal. A no-change wake remains silent. When ownership is ambiguous or the user asks who is working, respond with this compact synthesis: ```text Current task: <responsibility and scoped outcome> Internal workers: <owned implementation, review, QA, or audit roles; or none> External tasks: <request or evidence role only; ownership did not transfer; or none> Next: <current-task action or exact required decision> ``` Do not append raw cross-task messages, coordination backend events, heartbeat logs, worker transcripts, or claim telemetry. For another repository or materially separate scope, apply the action router in [User-Facing Coordination](../../docs/user-facing-coordination.md). Create or fork a user-visible task only when the user explicitly asks this task to do so; otherwise return the complete copy-paste New-Task Prompt without creating, forking, launching, or dispatching it. ## Closeout workflow ### 1. Establish the session scope Identify the task type: engineering, strategy, research, administration, or mixed. Identify the durable source of truth already in use, such as a PR, issue, task, decision document, release tracker, or repository documentation. Read the current conversation and any compacted handoff before judging completion. Do not restart completed work. ### 2. Verify current state Verify time-sensitive claims against the live source when tools are available. For engineering work, apply [Ordinary PR Closeout](../../workflows/pr-batch-integration-closeout.md#ordinary-pr-closeout) for evidence reuse, audit applicability, cleanup and archive readiness. Check the relevant subset of: - worktree cleanliness and intended commits; - local, remote, and PR head alignment; - PR state, required checks, reviews, unresolved threads, and merge status; - background agents, coordination claims, monitors, or active goals; - published artifacts or deployment state when those were part of the task. Preserve required check results and any escalation reason. When the wrapper reports delivery coverage, preserve its [report](../verify/references/verification-evidence.md#delivery-coverage): phase, candidate/base identities, selected/full coverage, required results, omissions, and selection reason. Selected integration success does not prove complete promotion checks. A retry or review stopping boundary leaves failed requirements open with disposition ownership; merge authority does not confer release/deployment permission. Apply the same reference's exact, clean-state rules before reusing local command evidence. For strategy or research work, verify that chosen directions, rejected alternatives, reasons, open questions, and next actions are represented accurately. If a material fact cannot be verified, label that exact fact `UNKNOWN`. Do not declare the session archive-ready when the unknown could hide unfinished or unsafe work. ### 3. Reconcile the loop Classify the session state: - **Done:** completed outcomes with verification evidence where available. - **Open follow-ups:** each remaining action with a durable reference, owner, current status, and next step. - **Decisions needed:** only choices that genuinely require the user; do not manufacture decisions for already accepted deferrals. - **Orphaned state:** uncommitted work, unpushed commits, unposted replies, live
From shakacode/agent-workflows, MIT.
More from shakacode/agent-workflows
- 0.592score
pause
shakacode
Print restart-safe copy-paste prompts for pausing an agent thread before restarting Codex, Claude, or another runner and for resuming in the same thread or a…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
pr-batch
shakacode
Plan and safely run one or more canonical issue, existing PR, or durably overridden ad-hoc work lanes with coordinated subagents, validation, review, and…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
fix-flaky-tests
shakacode
Use when a test fails intermittently in CI — classify the intermittency, work from the real CI error, fix the root cause systemically, and verify with a green…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
plan-pr-batch
shakacode
Use when choosing GitHub issues or PRs for a PR batch, recommending and grouping worker lanes by model/reasoning-effort assignment, preparing a subagent batch…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
post-merge-audit
shakacode
Use when auditing merged PRs after concurrent agent work, before a release candidate, after a suspected bad merge, or when checking for missed reviews…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
address-review
shakacode
Fetch GitHub PR review comments, triage them into must-fix/discuss/optional/skipped, and guide fixing or replying to selected feedback.
SkillBFreshness: B★ 8Claude Code +9 - 0.589score
audit-chats
shakacode
Audit multiple Codex tasks and visible ChatGPT chats for archive and follow-up actions, or recover interrupted Codex tasks when explicitly requested.
SkillBFreshness: B★ 8Claude Code +9 - 0.587score
evaluate-issue
shakacode
Use before fixing, batching, or assigning GitHub issues or proposed fixes when the value is uncertain, the report came from AI/code analysis, the fix is…
SkillBFreshness: B★ 8Claude Code +9 - 0.587score
plan-issue-triage
shakacode
Use when preparing a ready prompt for Claude, Codex, or another agent to perform a review-only GitHub issue triage or open-issue audit, especially broad scans…
SkillBFreshness: B★ 8Claude Code +9 - 0.587score
plan-review
shakacode
Use when reviewing an implementation plan before coding or launching workers to check approach, scope, and repo-convention fit.
SkillBFreshness: B★ 8Claude Code +9 - 0.587score
pr-walkthrough
shakacode
Explain a pull request as a conceptual walkthrough, either interactively one change at a time or as one complete COMMENT-only GitHub review when explicitly…
SkillBFreshness: B★ 8Claude Code +9 - 0.587score
pr-monitoring
shakacode
Use when monitoring an opened pull request through current-head checks, comments, conflicts, merge-readiness, and final handoff.
SkillBFreshness: B★ 8Claude Code +9
Signals
- Stars
- 8
- Forks
- 1
- Open issues
- 119
- Last commit (this folder)
- Sep 12, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/shakacode/agent-workflows
- Website
- shaka.shakacode.com
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from shakacode
- 0.592score
post-merge-audit
shakacode
Use when auditing merged PRs after concurrent agent work, before a release candidate, after a suspected bad merge, or when checking for missed reviews…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
plan-pr-batch
shakacode
Use when choosing GitHub issues or PRs for a PR batch, recommending and grouping worker lanes by model/reasoning-effort assignment, preparing a subagent batch…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
pr-batch
shakacode
Plan and safely run one or more canonical issue, existing PR, or durably overridden ad-hoc work lanes with coordinated subagents, validation, review, and…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
fix-flaky-tests
shakacode
Use when a test fails intermittently in CI — classify the intermittency, work from the real CI error, fix the root cause systemically, and verify with a green…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
pause
shakacode
Print restart-safe copy-paste prompts for pausing an agent thread before restarting Codex, Claude, or another runner and for resuming in the same thread or a…
SkillBFreshness: B★ 8Claude Code +9 - 0.592score
address-review
shakacode
Fetch GitHub PR review comments, triage them into must-fix/discuss/optional/skipped, and guide fixing or replying to selected feedback.
SkillBFreshness: B★ 8Claude Code +9
Cite this page
close-session skill. Agent Mart. https://agentmart.dev/skills/shakacode-close-session (last updated 2026-10-24)