Session anchoring protocol that prevents spec drift in long-running implementation sessions. Re-reads active spec, git state, issue state, and review comments to rebuild ground truth from source artifacts rather than relying on accumulated session context. Use when sessions exceed 30 minutes, after context compaction, before resuming paused work, when implementation feels misaligned with acceptance criteria, or when switching between tasks. Trigger with phrases like "anchor to spec", "re-read the spec", "am I drifting", "check alignment", "reload context", "what was I working on", "session too long".
npx claudepluginhub cianos95-dev/claude-command-centre --plugin claude-command-centreThis skill uses the workspace's default tool permissions.
Long sessions accumulate context that gradually diverges from the source of truth. This skill defines a re-anchoring protocol that rebuilds ground truth from artifacts rather than trusting session memory.
Verifies code implementations match specs, PRDs, epics, or tasks by checking completeness, acceptance criteria, edge cases, and scope creep. Use post- or during-implementation.
Verifies session work against agreed plan, creates issues for gaps and carryover tasks, runs quality gates, commits changes, mirrors to GitHub, and generates summary. Triggered by /close.
Guides Next.js Cache Components and Partial Prerendering (PPR) with cacheComponents enabled. Implements 'use cache', cacheLife(), cacheTag(), revalidateTag(), static/dynamic optimization, and cache debugging.
Share bugs, ideas, or general feedback.
Long sessions accumulate context that gradually diverges from the source of truth. This skill defines a re-anchoring protocol that rebuilds ground truth from artifacts rather than trusting session memory.
After 30+ minutes of implementation, agents commonly:
This is a direct consequence of treating agents as logic engines with perfect memory rather than orchestrators with finite attention. The anchoring protocol below rebuilds ground truth from source artifacts, not session memory. Source: Pierce Lamb Deep Trilogy — "treat Claude as an orchestrator with finite attention, not a logic engine."
Before executing this skill, gather the issue context bundle (see issue-lifecycle/references/issue-context-bundle.md). Include comment context in drift checks — comments may contain decisions that change scope, human overrides that alter acceptance criteria, or dispatch results that affect what remains to be done.
Before every task (or when triggered), re-read these four sources in order:
Read the PR/FAQ or spec document linked in the issue's linear frontmatter field:
exec mode, status, research readinessCheck the current implementation state:
git diff --stat since last commitRead the current issue from the project tracker:
exec:* mode, spec:* lifecycle stage, any blockersCheck for unresolved feedback:
/ccc:review output)| Trigger | Action |
|---|---|
| Session exceeds 30 minutes | Auto-suggest anchoring |
| Context compaction occurs | Mandatory anchor before continuing |
| Switching between tasks | Anchor to the new task's spec |
| After debugging a tangent | Anchor to verify you're back on track |
| Before claiming "done" | Final anchor to verify all criteria met |
Manual trigger (/ccc:anchor) | Full protocol execution |
After re-reading all sources, produce a concise alignment summary:
## Anchor Check — [Issue ID]
**Spec:** [spec title] | **Mode:** [exec mode] | **Status:** [spec status]
### Acceptance Criteria
- [x] Criterion 1 — implemented in [file:line]
- [ ] Criterion 2 — not yet started
- [~] Criterion 3 — partially implemented, [what remains]
### Drift Detected
- [Any misalignment between implementation and spec]
- [Any scope creep beyond acceptance criteria]
### Open Items
- [Unresolved review comments]
- [Open questions from spec]
### Next Action
- [The single most important next step]
/ccc:anchor runs the full protocol and produces the alignment summary output.exec:* mode. In exec:tdd, it also checks test state. In exec:checkpoint, it verifies gate status.