Skip to content
Agent Orchestration
Skill

/spectre-work-record

Record bounded history at owned Execute, Ship, standalone Create PR, blocked-handoff, snapshot or correction boundaries. Requires exact ownership. Not for reusable guidance, routine progress, batches or orchestrated Create PR.

BOOST
From plugin
spectre
16521 skills8 agents
Install
$ npx -y skills add joenandez/spectre --skill spectre-work-record --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/spectre-work-record

Context preview

The summary Claude sees to decide when to auto-load this skill.

Record bounded history at owned Execute, Ship, standalone Create PR, blocked-handoff, snapshot or correction boundaries. Requires exact ownership. Not for reusable guidance, routine progress, batches or orchestrated Create PR.

SKILL.md

spectre-work-record.SKILL.md
name: "spectre-work-record"
description: "Record bounded history at owned Execute, Ship, standalone Create PR, blocked-handoff, snapshot or correction boundaries. Requires exact ownership. Not for reusable guidance, routine progress, batches or orchestrated Create PR."
user-invocable: false

work-record

Purpose

Truthful work history; workflow state/telemetry own live progress. Capture never changes delivery, verification or acceptance authority.

Inputs

  • Exact run/work identity and applicable branch, PR or repository/base/head/diff association; observed lifecycle, account, entry points, tag intent and prior revision.
  • Owned boundary: one Execute start/completion; meaningful blocked handoff/resolution when authority, control or context prevents safe continuation/resumption; Ship after PR creation; standalone Create PR terminal write only without an Execute/Ship owner; explicit snapshot/correction, including outside workflows.

Working Set

  • Reuse applicable discoveries/loads; refine only unresolved questions. Read `references/work-capture-input.json` and `${PLUGIN_ROOT}/skills/spectre-capture/references/tagging-policy.md`.
  • Use acquired entry points; focus-read as needed: path, symbol/document section, observed positive line, role. Explain unavailable/non-code portions, including uncreated start locations; never invent locations or dump files.

Outputs + DONE

Short title/routing summary and required identity/provenance/status metadata, derived reliably. Complementary fields:

  • `requestedOutcome`: request; `scope`: boundaries.
  • `actualChanges`: delivered behavior/artifacts; `reasons`: non-obvious rationale/trade-offs. Start/blocked accounts distinguish delivered from undelivered.
  • `discoveries`: findings and maintained-authority links; `relatedContext`: typed entry points and bounded references.
  • `verification`: observed checks; `remainingWork`: residual implementation. Structured verification and PR lifecycle remain authoritative.

State absence truthfully; omit filler, duplicated narrative/state. New/revised accounts stay under 2,000 estimated rendered tokens; reference evidence, not logs.

**DONE:** permitted boundary has a truthful account or surfaced skip/no-op/conflict/failure, with exact work ID, revision and canonical `recordPath` when applicable.

Method / guardrails

1. One record per exact Execute run: resume revises it; distinct runs never fold regardless of branch/feature/candidate. Branches/PRs may reference multiple records. Execute-owned writes include `--source-run-id <exact-run>` alongside other identity flags for receipt derivation. 2. Execute start requires exact identity/branch; every not-yet-true section says `None yet.`, not placeholders (`TODO`, `REPLACE_ME`, angle brackets). Capture only the Inputs boundaries, never tasks, batches, checks, reviews, commits, ordinary decisions, routine progress, transient remediation or context refreshes. 3. New/current-run writes read `git rev-parse --abbrev-ref HEAD`, require explicit exact branch evidence, compare independent run-start evidence when available and preserve known same-run branch. Return recovery/skip if unavailable. Exact historical corrections may preserve unknown original branch by omitting `--branch`; never backfill from today's checkout. New historical snapshots require verified original branch. 4. Existing PR identity/URL requires `gh pr view <identity-or-url> --json state` outside store locks. Pass observed `pullRequest`; unavailable/unknown state, terminal-history contradiction or unqueryable draft without identity returns recovery/skip. Execution, verification and PR states stay separate:

  • `execution`: `unknown|in-progress|implementation-ready|acceptance-pending|blocked|finalized`.
  • `verificationState`: `unknown|not-run|checked|passed|failed`.
  • `pullRequest`: `unknown|none|draft-open|closed|merged`; bare `open` is invalid.

5. Authorized terminal Execute uses `execution.state: finalized`, `remainingWork: "None."` exactly (otherwise `CAPTURE_INPUT_INVALID`). Blocked/failed/interrupted accounts stay non-final with scoped residual work; review, CI, PR readiness, merge and closure are not implementation work. 6. Fill semantic JSON outside the store; submit on stdin: `knowledge-cli.mjs capture --kind work --input - --project-dir <project-dir> --json`. New/current-run writes add `--branch <exact-branch>` and exact identity via `--work-id <id>`, `--source-run-id <run>`, `--pull-request-id <pr>` or `--candidate <exact-json>`. Historical corrections require `--work-id <id> --expected-revision <revision>`. Changed records require loaded `revisionToken` as `--expected-revision` and preserve omitted tags; unchanged retries are no-ops. `--input <path>` is only for explicit manual/advanced capture or returned `recoveryInput`. 7. At owned terminal boundaries, promote only qualified reusable discoveries absent from maintained guidance via `$spectre:spectre-capture`, or link existing authority. No mandatory knowledge record per completion or added scan/write boundary. Capture failure/conflict returns recovery input without blocking delivery.

Handoff

Boundary, exact work ID, revision/conflict, canonical `recordPath`, associations, lifecycle facts and applicable recovery input.

Escalate-If

Unclear ownership/boundary, conflicting historical authority or unrecoverable write: surface recovery/skip without guessing or blocking delivery.

Read more
Ships withspectre

SPECTRE is an Agentic Engineering workflow for Claude Code and Codex designed to help you do ONE THING more, faster, and with higher quality.

Get the whole plugin

Other skills on spectre.