context-refresh
Skill40 starsChanged 3 days agoApache-2.0View on GitHub
Rebuild an accurate 0-to-1 picture of a workstream after time away — what was built, what is actually deployed right now, what is blocked, and what decisions are already settled. Use when returning to a branch after hours or days, when asked to "catch me up", "refresh my context", "where were we", "what's the status", "recap this thread", or before resuming work whose state you are reconstructing from memory rather than evidence. Also use before any status report to a founder, board, or team, because a briefing assembled from recollection is the highest-consequence place to be confidently wrong.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- Koroqe/claude-code-sdlctheir own version, first seen Oct 23, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- Koroqe/claude-code-sdlc★ 52same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository69 other files it gives its agents.
- AGENTS.mdAGENTS.md
- hushh-webapp/AGENTS.mdAGENTS.md
- impeccableSkill
- agent-orchestration-governanceSkill
- analytics-observability-governanceSkill
- animateSkill
- animation-vocabularySkill
- apple-designSkill
- autonomous-rca-governanceSkill
- backend-agents-operonsSkill
- backend-api-contractsSkill
- backend-runtime-governanceSkill
- backendSkill
- brandkitSkill
- client-env-paritySkill
- codex-skill-authoringSkill
- comms-communitySkill
- contributor-onboardingSkill
- docs-governanceSkill
- emil-design-engSkill
- find-animation-opportunitiesSkill
- founder-brief-curationSkill
- frontend-architectureSkill
- frontend-cache-coherenceSkill
- frontend-design-systemSkill
- frontend-native-surface-mapperSkill
- frontendSkill
- frontend-surface-placementSkill
- future-plannerSkill
- github-contribution-governanceSkill
- hushh-consent-mcpSkill
- hushh-icon-themeSkill
- iam-consent-governanceSkill
- imagegen-frontend-mobileSkill
- imagegen-frontend-webSkill
- improve-animationsSkill
- industrial-brutalist-uiSkill
- kai-test-account-resetSkill
- less-text-karma-jiSkill
- mcp-developer-surfaceSkill
- mobile-nativeSkill
- mobile-parity-auditSkill
- mobile-plugin-contractsSkill
- morphy-axSkill
- one-voice-governanceSkill
- oss-license-governanceSkill
- pdf-artifact-generationSkill
- pick-ui-librarySkill
- pkm-upgrade-rehearsalSkill
- planning-boardSkill
- pr-governance-reviewSkill
- profile-uiSkill
- prototypeSkill
- puppy-one-harnessSkill
- quality-contractsSkill
- release-ios-appstoreSkill
- repo-contextSkill
- repo-operationsSkill
- review-animationsSkill
- reviewer-app-testingSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 30
Use the complete sequence below after an absence or when reconstructing custody.
- Distinctive: line 34
Run these in order. Each one can invalidate your assumptions from the previous step, and the cheapest checks come first.
- Distinctive: line 48
Never state elapsed time from the conversation: If the user says "it's been 3 days", check — it may be five.
- Distinctive: line 62
Check the deployment history, not just the current state: who deployed last, when, and what SHA.
- Distinctive: line 85
Commit messages written at the time are more reliable than a summary written now.
- Distinctive: line 131
Burying the blocker: If something must happen before anything else can, it goes near the top, not in a closing list.
- Distinctive: line 133
Tidying away uncertainty: If a fact could not be verified, say so.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Context refresh
Apply the repo-wide Principal Craft Kernel and Bacterial Software Architecture Gate from
AGENTS.md; this skill adds context-recovery discipline, not authority to weaken
correctness, security, or verification.
This is verify-before-claim (.agents/skills/verify-before-claim/SKILL.md) pointed at a different
target. There, the risk is claiming a change works. Here, the risk is claiming you know
where things stand. Both fail the same way: a confident report assembled from memory.
The rule: reconstruct state from the live system, not from the conversation.
Anything the conversation told you was true was true then. Time has passed. main has
moved, deployments have been overwritten, PRs have gone stale, someone else has shipped.
A recap that replays the transcript is a recording, not a status report.
Quick status during active work
For an ongoing task, use the status brief. Refresh the
claims that may have changed; do not repeat a full repository or deployment audit
just to report progress. $context-refresh can produce this brief on request.
Use the complete sequence below after an absence or when reconstructing custody.
The refresh sequence
Run these in order. Each one can invalidate your assumptions from the previous step, and the cheapest checks come first.
1. Establish real time and real position
date -u '+%Y-%m-%d %H:%M UTC'
git fetch origin main --quiet
git log -1 --format='%h %ad %s' --date=short # what is actually at HEAD
git rev-list --count HEAD..origin/main # how far behind
git rev-list --count origin/main..HEAD # how far ahead
git status --short # uncommitted workNever state elapsed time from the conversation. If the user says "it's been 3 days", check — it may be five. Report the real interval; it changes how stale everything else is.
Divergence is the single most decision-relevant number after an absence. A branch that was current when you left may be hundreds of commits behind, which silently converts "ready to merge" into "needs a sync and conflict resolution first."
2. Verify what is actually deployed — never infer it
Deployment is the fact most likely to have changed while you were away, and the one people most often report from memory.
- Probe the live endpoints. Assert on response content, not just status. A
200that says "not found" is a soft failure that looks identical to health. - Check the deployment history, not just the current state: who deployed last, when, and what SHA. In a shared environment, someone else has probably deployed over you.
- Where a feature is meant to be dark, confirm it is still dark. A flag-gated route
returning
404is a positive result and worth stating as one.
Distinguish three separate claims that get conflated:
| Claim | What proves it |
| The code exists | It is committed on the branch |
| The code is deployed | A deploy run succeeded and the live service serves it |
| The feature works | A code path executed and produced the expected effect |
Most post-absence briefings quietly promote the first into the third. Say which one you actually have.
3. Reconstruct the arc from commits, not recollection
git log origin/main..HEAD --format='%h %ad %s' --date=short # what this branch adds
git diff origin/main...HEAD --shortstat # scale of the changeCommit messages written at the time are more reliable than a summary written now. Read them. Where this repo's practice has been followed they carry the reasoning, not just the change.
4. Check the open surfaces
Pull requests, CI status on the current head, and any background or scheduled work that may have completed or failed while you were gone. CI green on a SHA from last week says nothing about the SHA at HEAD today.
5. Separate settled decisions from open questions
Decisions the user already made are the most valuable thing to surface, because re-litigating them wastes the session and irritates the person who made them. Pull them out explicitly: constraints they set, options they rejected, and the reasoning where it was given.
Then state, separately, what is genuinely still open — and do not disguise an open question as a settled one to make the recap feel tidier.
Shape of a good refresh
Lead with orientation, then evidence, then direction:
- Where things stand — a short paragraph someone could act on immediately.
- What changed while you were away — often the single most useful section, and the one a transcript replay cannot produce.
- Deployment status — live-verified, with the environments that were deliberately left untouched named explicitly.
- The arc — why each piece exists, in the order it happened, since sequence usually carries the reasoning.
- Hard numbers — tests, diff scale, divergence, coverage gaps.
- Settled decisions — so they are not reopened.
- Where to pick up — including the first blocker, named plainly.
Match the medium to the volume. A few facts belong in the reply. A genuine 0-to-1 rebuild across many surfaces is easier to absorb as a structured document than as prose.
Failure modes this exists to prevent
- Replaying the transcript. The conversation is a record of the past, not a description of the present.
- Reporting stale deployment state. "It's deployed" was true five days ago; check.
- Silent staleness. Divergence from the trunk does not announce itself. Measure it.
- Confusing built with working. See the table in step 2.
- Burying the blocker. If something must happen before anything else can, it goes near the top, not in a closing list.
- Tidying away uncertainty. If a fact could not be verified, say so. An honest gap is more useful than a plausible guess, and far cheaper than one acted upon.
Pre-flight
[ ] I fetched the trunk and measured real divergence
[ ] I probed live endpoints and checked response content, not just status
[ ] I checked who deployed last and when, not just what is running
[ ] Elapsed time is measured, not quoted from the conversation
[ ] Every "deployed" or "working" claim maps to evidence I gathered this session
[ ] Settled decisions are listed so they are not reopened
[ ] The first blocker is stated near the top
[ ] Anything I could not verify is labelled as unverifiedDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

