Context catch up
Lightweight durable context for AI coding agents: 4 files, 9 rules, companion skills, and scripts that let agents self-iterate inside your workflow.
npx -y skills add lifan-builds/context-harness --skill context-catch-upAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Catch up a fresh agent session or true resumed session from context-harness files before planning or editing; do not use for ordinary mid-session follow-up turns after context is already loaded.
SKILL.md
3.4 KB, 699 tokens by cl100k_base, as published. Nobody here has run it
Context Catch-Up
Use this only at a fresh-session or true-resume boundary in a repository that already has context-harness files.
Repeat catch-up only after a real resume/compaction boundary, a materially
changed user objective, or an external context change. Otherwise use loaded
context, and use context-maintain only for updates, closeout, lesson capture,
or plan stress-testing.
Compatibility Check
Before ordinary catch-up, quickly inspect AGENTS.md, CONTEXT.md, and
NOW.md for:
<!-- context-harness:schema v3 -->in generated context files.- A small
AGENTS.mdContext Contract plus generated Context Index. - Required
CONTEXT.mdsections: Project, Structure, Operating Constraints, Workflow, Language, Relationships, Flagged Ambiguities, Learned Patterns. - A non-empty
NOW.mdwith current focus, blockers, immediate next step, and touched files.
If files are empty, partial, stale, or missing the v3 schema marker, report the
drift in the catch-up summary and route inspection or repair work to explicit
context-upgrade. Do not rewrite partial or schema-drifted layouts from
catch-up, and do not let non-blocking harness drift replace the user's requested
project task.
Read Order
- Read the small
NOW.mdrecovery packet first. - If the user supplied an objective, task, question, or relevant context, use
that as the hydrate query. If not, derive the query from
NOW.mdfocus and immediate next step. - Reconcile the user's context with
NOW.md: prefer the current user objective when scope changed, but surface material conflicts with repository state. - Run
node scripts/context-index.js hydrate "<query>"before openingPLAN.md, chunks, or bulky/task-specific context. - Read concise
CONTEXT.mddirectly only while it fits the startup budget; otherwise read hydrate-selected cards/sections. Even when it fits, keep the catch-up selective and do not load unrelated project history. - Use selected cards before large
PLAN.mdsections, chunks, or raw sources. - If there is no index, use only enough
CONTEXT.mdandAGENTS.mdfallback context for the current objective and route repair tocontext-upgrade. - If the project uses ADRs, skim only titles/dates relevant to the current task.
Output
Give the user a short catch-up summary:
- Current focus
- Active blockers
- Immediate next step
- Operating constraints or terms that matter for the requested task
- Any uncertainty that should be resolved before editing
Guardrails
- Do not rewrite context files during catch-up except to capture a durable
correction from the user; route layout repair to explicit
context-upgrade. - If
hydrate,check, or generated indexes fail or warn about stale generated files, mention the drift and use current markdown fallback context. Keep index repair as a follow-up unless the drift blocks correctness or safety. - Do not summarize raw external content into
CONTEXT.md; route task-local discoveries toPLAN.mdwhen work begins. - Keep the summary short enough that it helps the next action, not replaces it.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.