reconcile · diff
git:20260719.404ad2a to git:20260726.7992efa
59 added, 111 removed. Audit C to A.
---
- title: Reconcile Skill
- tags: [skill, cognitive, reconciliation, consistency, aigent-internal]
- aliases: [/reconcile, reconcile-skill]
- created: 2026-05-08
+ name: reconcile
+ agent: none
+ description: Read-only attention reconciliation. Computes actual seven-day project attention from daily notes against explicit Tier-1 intended shares.
+ allowed-tools: Read, Bash
+ user-invocable: true
+ status: PRODUCTION - deterministic and read-only
+ triggers:
+ - reconcile
+ - attention drift
+ - cross-system check
+ - consistency check
---
- # /reconcile — Cross-System Consistency Check
+ # /reconcile - attention reconciliation
- > [!info] aigent-OS-internal skill. Fires on /open (weekly cadence) or when Caddy detects drift signals. Proposes resolutions — does NOT auto-fix.
+ This measurement leg reports attention drift. It never edits priorities, daily
+ notes, goals, or runtime state.
- ## Invocation
+ Run:
- ```
- /reconcile
+ ```text
+ node daemons/nightly-reconcile.mjs --root <aigent-root> --as-of <fire-date>
```
-
- Optional: `/reconcile --full` to skip the 14-day stale threshold and check everything regardless of age.
-
- ---
-
- ## Step 1 — Load State Files
-
- Read all of the following in parallel. If a file is missing, note it and continue — absence is itself a finding.
-
- | File | Variable |
- |------|----------|
- | `$AIGENT_VAULT/memory/runtime/ACTIVE_STATE.json` | active objective, mode, focus |
- | `$AIGENT_VAULT/memory/runtime/GOAL_STACK.json` | active + queued goals |
- | `$AIGENT_VAULT/memory/runtime/BELIEF_STATE.jsonl` | current beliefs with confidence + last-checked |
- | `$AIGENT_VAULT/memory/runtime/SELF_MODEL.json` | recurring failure modes, strengths, calibration |
- | `$AIGENT_VAULT/memory/ACTIVE_PRIORITIES.md` | Tier 1–3 priorities |
- | `$AIGENT_VAULT/memory/runtime/LESSONS.jsonl` | recent lessons learned |
- | `$AIGENT_VAULT/memory/facts/facts.jsonl` | verified facts |
- | `$AIGENT_VAULT/memory/SKILL_GAPS.md` | open/resolved skill gaps |
- | `$AIGENT_VAULT/memory/DELEGATION_TRACKER.md` | active delegations |
- | `$AIGENT_VAULT/memory/runtime/BODY_STATE.json` | latest somatic capsule |
-
- ---
-
- ## Step 2 — Run Consistency Checks
-
- ### Check A: Goal/Priority Mismatch
- - Extract GOAL_STACK active goals (status: active)
- - Extract ACTIVE_PRIORITIES Tier 1 items
- - Flag any goal in GOAL_STACK with no corresponding Tier 1/2 priority
- - Flag any Tier 1 priority with no corresponding active or queued goal
- - **Mismatch = a goal is being tracked in one system but not the other**
-
- ### Check B: Stale Objectives
- - Compare ACTIVE_STATE `current_objective` field against GOAL_STACK active goals
- - If ACTIVE_STATE objective doesn't map to any active GOAL_STACK entry, flag as **orphan objective**
- - If the goal it maps to was completed/cancelled in GOAL_STACK, flag as **stale chase**
-
- ### Check C: Belief Contradictions
- - For each belief in BELIEF_STATE.jsonl, scan facts.jsonl for a fact on the same subject
- - If a fact directly contradicts a belief (opposite claim, same entity), flag as **contradiction**
- - If a belief has `last_checked` older than 14 days and `confidence < 0.7`, flag as **stale uncertain belief**
-
- ### Check D: Orphan Capsules
- - Read BODY_STATE.json for capsules with status: active or paused
- - For each capsule, check whether its `objective` field matches any active GOAL_STACK entry
- - If no match found, flag as **orphan capsule** (work in flight with no goal parent)
-
- ### Check E: Unresolved Skill Gaps
- - Read SKILL_GAPS.md for entries with `status: open`
- - Calculate age from `opened` date
- - Flag any open gap older than 7 days as **stale skill gap**
-
- ### Check F: Stale Delegation
- - Read DELEGATION_TRACKER.md for active items
- - Flag any delegation older than 14 days without a status update as **stale delegation**
- - Note the responsible agent and last update
-
- ### Check G: Self-Model Drift
- - Read SELF_MODEL.json `recurring_failure_modes` array
- - Read FAILURE_MODES.md for entries in the last 30 days
- - Flag any failure mode in SELF_MODEL that hasn't appeared in FAILURE_MODES.md in 30+ days as **possibly resolved** (self-model may be outdated)
- - Flag any failure mode appearing in FAILURE_MODES.md 2+ times recently that is NOT in SELF_MODEL as **unregistered pattern**
-
- ---
+ Use literal output and exit status as the checkpoint evidence.
- ## Step 3 — Output Reconciliation Report
+ ## Required intended-share input
- Print this exact format:
+ Read `vault/memory/ACTIVE_PRIORITIES.md` and locate explicit Tier-1 projects
+ with intended percentages. Accepted forms:
+ ```markdown
+ | Tier | Project | Intended share |
+ | 1 | [[Project Name]] | 40% |
```
- RECONCILIATION REPORT — {date}
- ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- Contradictions: {N}
- - {entity}: belief says "{belief_claim}", fact says "{fact_claim}"
+ or:
- Stale items: {N}
- - Skill gap "{gap_name}" open {X} days (threshold: 7)
- - Delegation "{item}" last updated {X} days ago (threshold: 14)
- - Belief "{belief_id}" not checked in {X} days, confidence {C}
+ ```text
+ Tier 1 - [[Project Name]] - intended 40%
+ ```
- Drift detected: {N}
- - Self-model lists "{failure_mode}" but no occurrence in last 30d (possibly resolved)
- - Failure mode "{pattern}" hit 2x recently but not in SELF_MODEL
+ Shares are integers from 1 through 100 and the Tier-1 total must be at most
+ 100. Names or aliases must resolve to a project note or an explicit canonical
+ name in the priorities file.
- Orphans: {N}
- - Capsule "{capsule_id}" objective has no matching active goal
- - Objective in ACTIVE_STATE has no GOAL_STACK entry
+ No intended shares returns nonzero:
- ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- Status: CLEAN ← if 0 issues
- Status: {N} issues found ← if >0
+ ```text
+ RECONCILE FAIL reason=intended-share-input-absent projects=0
```
- For each issue, append a **Suggested Action** line:
- ```
- → Suggested: {concrete action — update X, close Y, promote Z, verify W}
- ```
+ This is a configuration failure, never permission to invent shares.
- ---
+ ## Seven-day measurement
- ## Step 4 — Do NOT Auto-Fix
+ 1. Select valid `vault/daily/YYYY-MM-DD.md` notes inside the exact calendar
+ window `as_of-6d .. as_of`. A missing day contributes no mentions.
+ 2. Per project and note, count at most:
+ - 3 points for a heading or heading wikilink;
+ - 2 points for a body wikilink to a canonical note or alias;
+ - 1 point for an exact plain-text name.
+ 3. One occurrence receives only its highest applicable weight.
+ 4. Track project references outside Tier 1 as non-Tier-1 attention.
+ 5. Compute `actual = project points / all project points * 100`, rounded to one
+ decimal.
- This skill surfaces issues and proposes resolutions. It does not:
- - Modify any runtime file
- - Close any delegation item
- - Update ACTIVE_STATE or GOAL_STACK
- - Revise beliefs or facts
+ A zero denominator returns
+ `RECONCILE FAIL reason=no-measurable-project-attention`.
- The operator or the AIgent decides which resolutions to apply. Each fix should be deliberate.
+ ## Drift predicates
- ---
+ - Tier-1 drift: `actual < 0.40 * intended`.
+ - Combined non-Tier-1 drift: actual share over 25 percent.
- ## Caddy Enrollment
+ Output raw points, actual share, intended share, thresholds, drift decisions,
+ and source paths so the arithmetic is reproducible:
- Caddy fires this skill when:
- - `/open` runs and it has been 7+ days since last reconcile (check ACTIVE_STATE `last_reconcile` field)
- - A message contains "drift", "out of sync", "stale", "reconcile", or "consistency check"
+ ```text
+ RECONCILE PASS window=YYYY-MM-DD..YYYY-MM-DD notes=<N>
+ project=<name> points=<N> actual=<N.N>% intended=<N>% threshold=<N.N>% drift=<yes|no>
+ non_tier_1 points=<N> actual=<N.N>% drift=<yes|no>
+ ```
- See [[concepts/Cognitive Architecture Roadmap]] · [[memory/runtime/ACTIVE_STATE.json]] · [[memory/ACTIVE_PRIORITIES]]
+ Missing notes, shares, measurable attention, or parseable dates are not green.