docs(claude-code): onboarding overhaul + fix verify drift detection#37
Merged
Merged
Conversation
cmd_verify used `_load(LATEST) or snapshot(...)`, reusing the cached session-latest.json whenever it existed and ignoring the --live-context the /manifest command passes. Drift introduced after session start (a skill dropped in, a widened permission) was reported as "nothing added, nothing subtracted" against stale data, defeating the plugin's core promise. verify now always re-snapshots (merging live context) and saves the result, falling back to LATEST only when no snapshot is possible. Adds an end-to-end regression test driving cmd_verify against a baseline with post-baseline drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… loop Rewrite the plugin README so a Claude Code user gets the importance in the first screen: the composition (skills, permissions, MCP, instruction layer) is what decides what the agent can do, and it drifts silently. Adds a concrete threat list, a ~60s quickstart with the real first-session baseline message, and the everyday warning -> /manifest verify -> /manifest approve loop with real engine output. Keeps the honest "known limits" section. Polishes both command docs to match and to reflect verify's fresh re-snapshot. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Make the Claude Code integration easy for any Claude Code user to understand, add, and use, and fix a correctness bug the docs work exposed.
Docs / experience
warning -> /manifest verify -> /manifest approveloop with real engine output.policy_languagecomposite, hook-vs-command visibility).Fix (
verifydrift detection)cmd_verifyused_load(LATEST) or snapshot(...), so it trusted a cachedsession-latest.jsonand ignored the--live-contextthe/manifestcommand passes. Drift introduced after session start (a rogue skill, a widened permission) was reported as "nothing added, nothing subtracted" against stale data, defeating the plugin's whole point.verifynow always re-snapshots (merging live context) and saves the result, falling back to LATEST only when no snapshot is possible.exfilskill + a widenedBash(curl:*)permission on disk,verifynow reports both changes without a manual re-snapshot.Validation
python -m pytest claude-code/tests/test_capture.py -q-> 7 passed (adds an end-to-end regression test drivingcmd_verifyagainst post-baseline drift).ruff check claude-code/engine claude-code/tests --select E,F,W --ignore E501-> clean.snapshot/approve/verifyend-to-end against a synthetic~/.claude.Notes
integration.yamlclaimstested_against: agent-manifest 0.2.0; PyPI now has 0.3.0 for bothagent-manifestandagentrust-trace. Left version numbers untouched (no maintainer test matrix re-run in scope); flagging for a follow-up.🤖 Generated with Claude Code