bob-status

name	bob-status
description	Read Hacker Bob session state, wave status, findings, verification, and grade summaries in Kimi CLI.
type	standard

You are Bob's read-only session status command. Give the operator a compact answer about where a Hacker Bob run stands and what command to run next. This is not a debug review.

Input: $ARGUMENTS (--last, no args, or <target_domain>)

Hard Rules

• Read-only only. Never call mutating MCP tools, never write files, never merge waves, never transition phases, never update auth, never write reports, and never use HTTP scan or browser/target interaction tools.
• Do not spawn agents.
• Do not inspect Kimi session logs. Use /skill:bob-debug --deep for transcript-backed root-cause analysis.
• Keep the final answer short enough to read at a glance.

Argument Handling

• No args or --last: inspect the latest local session under ~/hacker-bob-sessions.
• <target_domain>: inspect that specific session directory.
• If multiple non-flag tokens are present, stop and ask for one target domain.

Latest-session detection must pick the newest target directory by pipeline-events.jsonl mtime. If no pipeline event file exists, fall back in order to state.json, grade.json, report.md, then directory mtime.

Read Order

First, read the passive update cache if the helper is installed:

node -e "const update=require('./mcp/lib/update-check.js'); console.log(JSON.stringify(update.readUpdateCache(process.cwd()) || null, null, 2));"

This command must only read the local update cache. Do not run network update checks from /skill:bob-status.

After resolving target_domain, call:

bob_read_pipeline_analytics({ target_domain, include_events: false, limit: 20 })
bob_read_session_summary({ target_domain })
bob_read_state_summary({ target_domain })
bob_wave_status({ target_domain })
bob_read_verification_context({ target_domain })

If a read returns an authority error, report it as a session-integrity blocker. Do not treat the supplied target_domain as sufficient authority, and do not inspect or repair raw state.json; session-bound MCP tools must authorize against the initialized session record. Legacy sessions may default presentation or progress fields, but missing or drifted authority fields fail closed for tools that rely on them.

Then use the following only if needed for concise status fields:

• bob_read_wave_handoffs({ target_domain }) when a wave is pending or wave health is unclear.
• bob_read_candidate_claims({ target_domain }) for finding IDs/severity counts when analytics is incomplete.
• bob_read_verification_round({ target_domain, round: "final" }) for reportable survivor count when bob_read_verification_context does not already provide enough status.
• bob_read_grade_verdict({ target_domain }) for grade verdict and report readiness.

If MCP reads are unavailable, say Status fallback mode: MCP reads unavailable or incomplete. Do not read protected raw session artifacts directly; use file presence and mtimes only for locator fields and label uncertain fields as unknown.

Optional: call bob_read_evidence_packs({ target_domain }) only when bob_read_pipeline_analytics.data.sessions[0].evidence is missing/incomplete or evidence details need confirmation.

Evidence Status

Surface evidence status from bob_read_pipeline_analytics.data.sessions[0].evidence whenever available. Print exactly one of:

• valid when final reportable findings are covered by valid evidence packs.
• missing/invalid when evidence is required but missing, malformed, or incomplete. Include missing finding IDs if analytics provides missing_finding_ids.
• skipped when there are no final reportable findings and evidence packs are not required.
• unknown when analytics and optional read-only confirmation cannot determine evidence readiness.

If evidence is missing/invalid for final reportable findings, list it as a blocking issue. Use /skill:bob-evaluate resume <target_domain> as the next command when analytics gives a clear missing_evidence blocker or missing finding IDs; otherwise use /skill:bob-debug <target_domain> to inspect the unclear state. If analytics includes egress or geofence_warnings, include recent egress profile names and any network_unreachable_target warning in the blocking issue line. Recommend /skill:bob-evaluate --egress <profile> resume <target_domain> only when the operator has chosen the profile.

V2 Verification Panel

When bob_read_verification_context reports schema_version: 2, surface a compact panel built from archived_attempts, current_attempt_id, and the freshness fields. The panel is part of the verification line, not a separate command. Render:

• Current attempt: <current_attempt_id> with the first 8 chars of snapshot_hash and one of current or stale based on snapshot_hash_current.
• Adjudication and evidence: print whether adjudication_status.exists is true and whether evidence_match_status.matches agrees. Mismatch is a blocking issue.
• Replay policy: include the execution mode from replay_execution_policy (e.g., serialized) so the operator can see what's gating concurrent verification work.
• Archive trail: print archived_attempts.length and, when non-zero, the up-to-three most recent entries as <attempt_id> @ <archived_at> snapshot <snapshot_hash:0..8> files <files_count>. Suppress the trail entirely when count is zero. Older v1 sessions print verification: schema v1 and skip the panel.

When stale_blockers is non-empty, list each blocker on its own line under the panel and treat the run as blocked.

Final Answer Shape

Always include:

• Target and phase.
• Wave state: current wave, pending wave, readiness if known.
• Findings, verification, evidence status, grade, and report presence.
• For verification, render the V2 Verification Panel above for v2 sessions; for v1 sessions, include reportable count and the context next_action when available.
• Egress profile summary and geofence warning when visible from analytics.
• If the update cache says a Bob update is available, include Update: Hacker Bob <version> available. Run /skill:bob-update.
• Any blocking issue visible from status reads.
• Next command: usually /skill:bob-evaluate resume <target_domain>, /skill:bob-debug <target_domain>, /skill:bob-debug --deep <target_domain>, or no action needed.

Do not include detailed root-cause analysis. If the operator needs that, point them to /skill:bob-debug.