CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

ledger

Sweep the full current session, repo/task context, local task files, and (with approval) connected remote task stores for outstanding actionable work, then record only useful high-confidence items in docs/outstanding-issues.md with priority, context, and next steps. Use when the user types "/ledger" or asks to extract outstanding tasks from this session into the ledger.

Quality

89%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

ledger — extract outstanding work from this session into the universal task ledger

/ledger is a session sweep: it mines everything visible to the current task for outstanding actionable items and records the high-confidence ones in docs/outstanding-issues.md — the same universal ledger the /issues skill owns. /ledger is the extractor; /issues remains the reader/editor. Follow every writing rule in the issues skill.

Disambiguation: this skill targets the outstanding-issues ledger only. It must never read or write docs/branch-review-ledger.md, and the ledger:lookup / ledger:index npm scripts belong to that review history — they are not part of this skill.

1. Sweep (read-only)

Collect candidates from, in order:

  1. Full chat context — every request, finding, decision, error, and deferral in this session.
  2. Repo/task context — current branch, git status, uncommitted/unpushed work, recent commits on this branch, and TODO/FIXME markers in files touched this session (not a repo-wide grep).
  3. Local task context files — docs/outstanding-issues.md (baseline for dedupe), docs/process-hardening.md debts, session scratchpad notes, and any plan/handoff notes this task produced.
  4. Remote task storage — GitHub PRs/issues, Todoist or other connected MCP task stores. These are provider-backed: read them only with explicit user approval per the AGENTS.md provider boundary. If not approved or not connected, skip and say so in the report.

Hunt specifically for: deferred or "later" items; unresolved "need to", "should", "TODO", "follow up", "please do", "investigate", "fix" statements; explicit blockers and errors left unfixed; and acceptance criteria stated but not met.

2. Report before writing

Present findings in this order, then pause for confirmation on anything ambiguous:

  1. Confirmed outstanding tasks — high confidence, clearly actionable.
  2. Risky / blocked areas — items blocked on approvals, providers, or user decisions.
  3. Quick skill suggestions — up to 3 ideas to improve /ledger itself (report only; add to the ledger only if the user confirms).

Render each candidate with the full structured record:

id: short-slug                      # e.g. ledger-missing-trigger
title: one line
importance: P0 | P1 | P2 | P3
category: bug | follow-up | risk | refactor | infrastructure | hygiene | idea
why: business value / user impact
context: what in this chat triggered it
owner: assistant | user | unknown
next_step: concrete next action
dependencies: ids or #NNN rows, if any
status: pending | done | blocked
confidence: high | medium | low
suggested_improvements: optional related idea (omit if none)

3. Filter and dedupe

  • Record only useful, high-confidence items. Drop vague observations, restated repo policy, and anything already fixed this session.
  • Dedupe against existing open rows by intent, not exact wording. Same intent → do not re-add; report it as "duplicate of #NNN". New information on an existing row → npm run issues:update instead of a new row.
  • Medium/low-confidence items: list them in the report, add only the ones the user confirms.

4. Write

Queue each retained item as an immutable inbox request; never edit the canonical ledger:

npm run issues:add -- --pri P2 --type issue --summary "<title>" --detail "<detail>" --source "session YYYY-MM-DD /ledger sweep"

Field mapping into the request payload (the reconciler later creates the canonical row):

  • importance → Pri: P0 and P1 → P1 (note "was P0" in Detail), P2 → P2, P3 → P3.
  • category → Type: bug/risk → issue; follow-up/refactor/infrastructure/hygiene → task; idea → rec.
  • Detail = next_step, then Why:, Context:, Owner:, Confidence:, and Depends on: as short clauses. Escape | as \|. Keep it one crisp cell, not a paragraph.
  • Source = session YYYY-MM-DD /ledger sweep plus the doc/PR/file:line when one exists.

If no new items survive the filter, create no request and do not change the canonical ledger.

After queueing, run npm run check:outstanding-issues and paste its decisive line. Do not refresh the visual register because the canonical ledger is unchanged. If the user explicitly asks for a commit, commit only the newly created request file(s); never push unless the user asks or a handoff is already in flight. A dedicated fresh-base branch later runs npm run issues:reconcile.

Fallback (other workspaces): if docs/outstanding-issues.md does not exist, search for similarly named files (outstanding*, *issues*, TODO.md, BACKLOG.md); if none fit, propose a location and ask before creating. A newly created ledger uses concise structured Markdown bullets carrying the full record above, appended chronologically, preserving existing entries.

Boundaries

  • No provider-backed or live-service reads/writes without explicit approval.
  • Recording an item that touches a protected surface (RAG ranking, clinical, privacy) is fine; acting on it later still needs that surface's own gate and flagging rules.
  • /ledger never deletes or rewrites existing rows — cleanup only on explicit request, via /issues.
Repository

BigSimmo/PsychSift

Path

.claude/skills/ledger/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.