Skip to content

mataeil/OODA-loop / skills/ooda-status

ooda-status

Verified

Display OODA-loop status dashboard. Shows cycle count, domain states, confidence scores, action queue, and alerts in a single view.

/ooda-status — Status Dashboard

Display the current state of the OODA-loop at a glance.

Step 0: Safety Check

Check for the HALT file before anything else.

ls agent/safety/HALT 2>/dev/null && echo "HALT_ACTIVE" || echo "HALT_INACTIVE"

If HALT exists, continue rendering the dashboard but mark HALT status as "ACTIVE". Do not abort — status must always be readable regardless of HALT state.

Step 1: Gather Data

Read each file below. If a file is missing, note it and use a safe default value. If a file exists but contains invalid JSON (parse error), treat it the same as missing and append an alert: [WARN] Corrupt state file: <path> — skipped (parse error). Do not let a single corrupt file abort the entire dashboard.

config.json — project name, current level, cost limit, domain list, domain status fields

cat config.json 2>/dev/null || echo "MISSING"

Also read config.mission (the project's purpose the loop drives toward) and each domain's mission_alignment.

Read each domain's status field from config.json to identify which domains are "available" (not yet configured) vs "active" vs "disabled".

agent/state/evolve/state.json — cycle_count, last_cycle timestamp

cat agent/state/evolve/state.json 2>/dev/null || echo "MISSING"

agent/state/evolve/confidence.json — per-domain confidence scores (0.0–1.0)

cat agent/state/evolve/confidence.json 2>/dev/null || echo "MISSING"

agent/state/evolve/action_queue.json — pending and proposed action counts, top action

cat agent/state/evolve/action_queue.json 2>/dev/null || echo "MISSING"

agent/state/evolve/metrics.json — execution counters/streaks (under counters)

cat agent/state/evolve/metrics.json 2>/dev/null || echo "MISSING"

Cost source: today's spend comes from agent/state/evolve/cost_ledger.json (total_estimated_usd, after confirming date == today UTC; stale date ⇒ $0.00 pending reset) and the limit from config.cost.daily_limit_usd. metrics.json has NO cost fields — never read cost from it.

Domain state files — for each domain in config.domains, read its state_file path:

  • last_run timestamp
  • status field (healthy / degraded / critical / error)
  • score value
  • alerts array
# Example: cat agent/state/health/state.json 2>/dev/null

Step 2: Calculate Display Values

Time formatting — compute elapsed time from last_run or last_cycle to now:

  • < 60 minutes → Xm
  • < 24 hours → Xh
  • = 24 hours → Xd

  • Never run → —

Domain status symbol:

  • ✓ if status is healthy
  • ⚠ if status is degraded
  • ✗ if status is critical or error
  • ? if domain has never run or state file is missing

Score — numeric value from domain state, formatted to 2 decimal places. Show — if unavailable.

Confidence — value from confidence.json for this domain (e.g. 0.9). Show — if unavailable.

Actions — count items in action_queue.json by status field: pending vs proposed. Top action: first item sorted by RICE score descending.

Alerts — collect all alerts arrays from domain state files. Count total. Show none if count is 0, otherwise show the count.

Cost — cost_ledger.json.total_estimated_usd (if date == today UTC, else $0.00) / config.cost.daily_limit_usd. Show —/— if unavailable.

Orient Health (v1.2.0) — parsed from the Orient layer state files:

  • episodes_count and last_episode_week — from episodes.json.episodes[]. Show 0 and — if empty or missing.
  • principles_count and principles_high_conf — from principles.json.principles[], where high_conf is count(confidence >= 0.5).
  • lens_domains — number of agent/state/*/lens.json files that exist; denominator is the count of active domains in config.domains.
  • chain_count_last_10 — in state.json.decision_log[-10:], count entries where chain_executed exists and is non-empty.
  • active_interventions — length of memos.json.interventions[].
  • skill_gaps_unaddressed — count(gap.resolved != true) in skill_gaps.json.gaps[]. Break out learning_loop_break count separately since those flag internal evolve invariants (e.g., cost_ledger auto-patches).
  • reflections_count and last_lesson — from reflections.json.reflections[]: total count and the most recent entry's lesson (truncate to ~32 chars). Show 0 / — if the file is empty or missing. This shows whether the Reflexion self-critique loop (evolve Step 5-F / 2-F) is actually running.

Season + Focus (v1.2.0) — from config.json:

  • season_mode — config.season_modes.current_mode if enabled, else "disabled".
  • season_overrides_count — length of config.season_modes.modes[current_mode].weight_overrides.

Active context (v1.2.0) — from config.json.active_context.path:

  • if set and file readable: <path> (age: <mtime>m); else none.

Step 3: Render Dashboard

Print the dashboard using box-drawing characters exactly as shown below. Replace {placeholders} with the computed values from Step 2.

╔══════════════════════════════════════════════════════╗
║  OODA-loop status                                    ║
║ Mission: {mission_oneline_or_—}                       ║
╠══════════════════════════════════════════════════════╣
║ Cycle: #{N}  Last: {ago}  Level: {N}  Vel: {N}/day   ║
╠══════════════════════════════════════════════════════╣
║ Domain           Score  Conf  Trend  Last  Status    ║
║ {domain_name}    {score} {conf} {↑↓→}  {ago} {sym}   ║
╠══════════════════════════════════════════════════════╣
║ Actions: {N} pending, {N} proposed  Oldest: {age}d   ║
║ Next: {top_action_title} (RICE {score})              ║
╠══════════════════════════════════════════════════════╣
║ Saturation: {N} observe-only cycles {bar}            ║
║ Alerts: {count_or_none}                              ║
║ HALT: {ACTIVE / inactive}                            ║
║ Cost: ${spent}/${limit} today (${rate}/h)             ║
╠══ Orient Health (v1.2.0) ═══════════════════════════╣
║ Episodes: {episodes_count} (last: {week})            ║
║ Principles: {principles_count} ({high_conf} conf≥0.5)║
║ Lens: {lens_domains}/{active_domain_count} domains   ║
║ Chain exec: {chain_count_last_10}/10 cycles          ║
║ Interventions: {active_interventions} active         ║
║ Gaps: {skill_gaps_unaddressed} ({loop_break} break)  ║
║ Reflections: {reflections_count} (last: {last_lesson})║
╠══ Season + Context (v1.2.0) ════════════════════════╣
║ Season: {season_mode} ({overrides_count} overrides)  ║
║ Context: {context_path_or_none}                      ║
╚══════════════════════════════════════════════════════╝

The --orient flag opens a detailed view of Orient Health only (useful when debugging whether the learning loop is actually running):

/ooda-status --orient

Output focuses on Episodes / Principles / Lens / Chain / Interventions / Skill gaps and omits the domain/cost/saturation rows.

--scorecard — is the loop actually WORKING?

/ooda-status --scorecard          (all recorded cycles)
/ooda-status --scorecard --window 20   (last 20 cycles)

Renders the Loop Scorecard — the loop-engineering measurement view that answers "is the loop improving the project, or just running?". This is the deterministic reference scripts/loop_scorecard.py rendered verbatim; it reads outcomes.json (Step 6-C9), metrics.json counters, cost_ledger.json, and action_queue.json — no recomputation, no model call. KPIs (the measurement canon):

  • Loop Value Score — mean quality_multiplier across scored cycles (0–1). The single headline number: 1.0 = every cycle merged & held; 0.0 = all futile.
  • Task Completion Rate — % of cycles that merged and were accepted.
  • Futile Cycle Rate — % of cycles that ran but changed nothing.
  • PR Merge Rate + hold rate (merged that weren't reverted).
  • Action Queue Resolution — resolved ÷ added (a value < 100% means the backlog is growing faster than the loop clears it).
  • Cost per Successful Cycle — total cost ÷ accepted-value cycles.
  • Goal Progress — mean progress of active goals.json done-conditions (the loop-engineering "run until a verifiable goal is met" signal).
  • Gap Resolution / Lesson Application — learning-loop health: are self-diagnosed skill gaps closed, and are reflexion lessons re-applied?
  • Verdict — working / partial / stalled, from the Loop Value Score.

Graceful degradation: with no outcomes.json yet (pre-v1.4.0 state or a fresh project), every KPI shows — and the verdict reads "no outcomes recorded yet."

--share — render the latest Cycle Card

/ooda-status --share re-renders the most recent cycle's Cycle Card — the same shareable artifact evolve prints at the end of its Step 7 — so it can be screenshotted or pasted into X / Reddit / Slack without re-running a cycle. It is read-only.

/ooda-status --share

--share --plain — emit only the text line

If the --plain flag is appended (/ooda-status --share --plain), this mode acts as a strict subset of --share that omits the Cycle Card rendering output. It reuses identical state reconstruction logic, LEARN-line selection priority, and all graceful degradation rules defined in --share. It emits only the single-sentence plain-text share line (as defined in evolve Step 7).

/ooda-status --share --plain

Reconstruct the card (or plain text) from existing state — no recomputation:

  • header / DECIDE / ACT — state.json.decision_log[-1] (cycle, timestamp, domain, skill, score, confidence, result, pr_number, risk_tier, orient_summary).
  • OBSERVE / ORIENT — decision_log[-1].orient_summary and, if present, the **Orient** line of the latest entry in agent/state/evolve/CHANGELOG.md.
  • LEARN — pick the highest-signal change using the SAME priority order as evolve Step 7 (human-decision confidence change > lens change > new intervention > micro-adjustment). Source it, in order, from the latest agent/state/*/lens_changelog.json entry, memos.json.interventions[] created this cycle, and the **Confidence** (trend / micro-adj) line of the latest CHANGELOG.md entry. If none is recoverable, render no new orientation recorded for cycle #{N}.
  • COST — latest cost_ledger.json entry + config.cost.daily_limit_usd.

Render byte-for-byte the same box and the plain-text share line as evolve Step 7, including the honesty rule on verbs (re-aimed / adjusted / deprioritized — never "trained" or "learned weights") and the same missing-field graceful degradation (render — for any absent field; on legacy pre-v1.2.0 state expect more —). If no cycle has run yet (decision_log empty), print: No cycle to share yet. Run /evolve first.

New columns and rows explained:

  • Vel (Velocity): cycles per day = total_cycles / days_since_first_cycle. Helps detect runaway loops or idle periods.
  • Trend: per-domain confidence direction. decision_log only snapshots the WINNER's confidence each cycle, so compare the domain's two most recent decision_log appearances (↑/↓/→ by delta); a domain with fewer than two appearances in the retained log renders →. Do not pretend a "5 cycles ago" per-domain snapshot exists — it doesn't.
  • Oldest: age of the oldest pending action in action_queue, in days. Shows — if queue is empty. Highlights aging items that may need human review.
  • Saturation: consecutive_observe_only_cycles from state.json. Render as a progress bar toward saturation.halt_threshold (e.g., ████░░░░░░ 40%). Shows 0 if no saturation.
  • $/h (Cost rate): cost_ledger.total_estimated_usd / hours_since_midnight_utc. Helps predict whether daily limit will be hit.

One row per enabled domain. Pad domain names and numbers so columns align. If HALT is active, write HALT: ACTIVE (all caps, no color codes needed -- emphasis via caps).

Narrow terminal fallback — The box-drawing layout above assumes >= 50 columns. If the output environment is narrow (e.g. split pane, mobile terminal), fall back to a compact plain-text list without box-drawing characters:

OODA-loop status
Cycle #0 | Last — | Level 0
---
service_health  — — — ?
test_coverage   — — — ?
---
Actions: — pending, — proposed
Alerts: none | HALT: inactive | Cost: —/—

Step 4: Suggestions

After rendering the dashboard, check whether a suggestion should be shown:

  1. Read cycle_count from state.json. If cycle_count >= 3, proceed.
  2. Collect all domains from config.json where status: "available".
  3. If any "available" domains exist, show exactly one suggestion per status check:
  Suggestion: You've run {N} cycles. Consider adding
  /scan-market for strategic insights.
  Run: /ooda-skill create scan-market

Replace the domain name and description with the actual available domain being suggested.

  1. Rotate through available domains across successive status checks (use cycle_count mod available-domain-count to pick which one to show).
  2. Once all domains are either "active" or "disabled" (none remain "available"), omit the Suggestions block entirely.
  3. Never show more than 1 suggestion per status check.

Graceful Degradation

ConditionBehavior
config.json missingPrint: Not configured. Run /ooda-setup first. — stop.
state files missing (all)Show dashboard with Cycle: #0 Last: — Level: 0 and domain rows as ?. Add note: No cycles run yet. Run /evolve to start.
Individual domain state file missingShow ? for score, conf, last, status for that domain only.
Any state file contains invalid JSONTreat as missing (use defaults), add [WARN] Corrupt state file: <path> to alerts section.
action_queue.json missingShow Actions: — pending, — proposed and Next: —.
cost_ledger.json missingShow Cost: —/— today.
HALT activeShow full dashboard. Mark HALT: ACTIVE. Do not suppress any data.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall