ground-meta-project
Skill0 starsChanged yesterdayMITView on GitHub
Context recovery and drift diagnosis skill: restores goal, state, identity and authority after session changes, moves, updates or compaction; diagnoses context drift (unconfirmed hypotheses, contradictions, authority overreach) without automatic rollback.
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- templates/AGENTS.mdAGENTS.md
- decommission-meta-projectSkill
- init-meta-projectSkill
- manage-meta-projectSkill
- record-observationSkill
- scan-meta-projectSkill
- sync-architectureSkill
- tune-meta-projectSkill
- update-meta-projectSkill
- decommission-meta-projectSkill
- ground-meta-projectSkill
- init-meta-projectSkill
- manage-meta-projectSkill
- record-observationSkill
- scan-meta-projectSkill
- sync-architectureSkill
- tune-meta-projectSkill
- update-meta-projectSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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 40
Read ONLY what the situation needs, by semantic section reference:
- Distinctive: line 72
only a technical plan re-draft is needed within existing authority → do it.
≈ “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
Agent Grounding: Context Recovery & Drift Diagnosis
When to invoke skill
Two modes:
- Recovery: new session, repository moved, task handoff, after
update-meta-project, after context compaction. - Drift diagnosis: repeated unconfirmed hypotheses, self-contradictions, actions beyond authority, success criteria adjusted to pass checks.
Semantic triggers only (a normal iterative TDD loop is NOT a violation): repeatedly failing edits on one defect, new adjacent regressions, or the user explicitly calling for a zoom-out.
Execution steps
Phase 0: Mode & Stop
- Determine the mode from the invocation context.
- Drift mode: FIRST stop all modifications, snapshot
git status -s/git diff --stat, and only then inspect. NEVER roll back automatically; unknown authorship of changes is reported, not reverted.
Phase 1: Identity (assignment + validation)
- The assigned role comes from the session assignment or direct user instruction — this skill never reassigns it.
- Validate it mechanically:
python scripts/verify_identity.py --actor <actor_id>Admission requires an ACTIVE binding on an ACTIVE installation; an unresolvable identity is reported as unknown, never guessed from directories or foreign passports.
Phase 2: Selective Context Recovery (read-only)
Read ONLY what the situation needs, by semantic section reference:
- passport: North Stars + Local Verification Tooling;
- the agreed scenario from
.agents/meta-project.json; - the live task: the CURRENT user request first;
tasks.mdsecond — a fresh request wins over a stale entry, but never silently cancels previously agreed restrictions; contradictions are surfaced explicitly; - relevant fragments of
architecture.md/boundaries.md; - the assigned role's OWN
notes.md(foreign passports are not loaded); - situation-relevant reports and observations (a found report is context, NOT an authorization). Prior-session chat is assumed unavailable; if the original task is not recoverable from the sources above, state the absence.
Phase 3: Diagnosis (drift mode)
Compare the working tree against the recovered context: boundary drift, symptom masking vs root cause, data-flow tracing (Origin ➔ Transport ➔ Persistence ➔ Consumer), contract duplication. Separate FACTS from HYPOTHESES (a file in the diff does not prove who changed it).
Phase 4: Compact Result & Conditional Continuation
Output exactly this structure:
Goal and completion criterion:
Task meaning & preserved invariants:
Environment & identity:
Established state:
Discrepancies & unknown facts:
Next step & how it will be verified:
CONTINUE / DECISION NEEDED: <if decision — ONE concrete question>Then continue conditionally:
- conditions of already-authorized work restored → CONTINUE it;
- only a technical plan re-draft is needed within existing authority → do it;
- a new choice, authority extension, or substantive conflict → ask ONE concrete question;
- an explicit user pause request always wins.
Grounding itself stays read-only: no telemetry rows, no tasks.md changes, no lock files, no persistent reports.
Does 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.

