agentsclimarketplace

Context catch up

Skill lifan-builds/context-harness/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.

Install
npx -y skills add lifan-builds/context-harness --skill context-catch-up

Assembled 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.md Context Contract plus generated Context Index.
  • Required CONTEXT.md sections: Project, Structure, Operating Constraints, Workflow, Language, Relationships, Flagged Ambiguities, Learned Patterns.
  • A non-empty NOW.md with 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

  1. Read the small NOW.md recovery packet first.
  2. If the user supplied an objective, task, question, or relevant context, use that as the hydrate query. If not, derive the query from NOW.md focus and immediate next step.
  3. Reconcile the user's context with NOW.md: prefer the current user objective when scope changed, but surface material conflicts with repository state.
  4. Run node scripts/context-index.js hydrate "<query>" before opening PLAN.md, chunks, or bulky/task-specific context.
  5. Read concise CONTEXT.md directly 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.
  6. Use selected cards before large PLAN.md sections, chunks, or raw sources.
  7. If there is no index, use only enough CONTEXT.md and AGENTS.md fallback context for the current objective and route repair to context-upgrade.
  8. 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 to PLAN.md when 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.

Keep looking

Skills are one crate of 326,984. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.