David F GliddenandClaude Opus 5
3af7207c29
The wrap was complete and committed; the steward then ran /doctor, which found and fixed real breakage. Recorded as an addendum to the same session file rather than a new one — it is the same session, past midnight. Fixed: 8 sub-agent name collisions across 21 files (15 renamed; 0 remain, 48 unique names — confirmed by the harness, which surfaced 15 previously invisible agents immediately). 4 invalid SKILL.md frontmatters, latent rather than live since the harness parses leniently. plane MCP disabled (0 calls in 41 real sessions). permissions.defaultMode set to auto. chamber-library/CLAUDE.md trimmed 54,129 -> 22,530 chars, under the warning threshold, all 42 tool names and all 9 sections preserved. TWO LOOSE ENDS recorded in ADDENDUM 1 and flagged in MEMORY.md, because either would strand the next session: 1. chamber-library/CLAUDE.md is UNCOMMITTED — a 31,861-char deletion, steward-approved at the doctor gate, left for the steward because the doctor protocol forbids the executor committing CLAUDE.md edits. git diff --stat reports 11 lines and badly understates it; the cut sections were single 16k/20k-char lines. 2. ~/.claude/agents is NOT GIT-TRACKED — 51 hand-edited files, 0 tracked, not a symlink into dotfiles. The renames therefore have no version history, and the only backup went to a session-scoped scratchpad that dies on clear. The full rename mapping is written into ADDENDUM 1 and is the only undo that survives. The wider gap is noticed and NOT filed, per the proposed moratorium: a governed surface with no version control, in a system whose premise is that the record must be checkable. Found by accident. Also logged: a thirteenth instrument error. I reported dotfiles as having unpushed commits to "origin" by conflating two repos' status lines — dotfiles has gitea and github and no origin; the origin line was studium-engine's. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017vKkg2EJF1rGwdFdBogwqx
5.8 KiB
name, description
| name | description |
|---|---|
| landscape-scan | Survey the competitive/comparable-project landscape for CapableMind every couple of days — projects similar or too-identical to what we're building. Generative, not anxious: learn, borrow, sharpen our bet, and notice the day someone builds the governed/epistemic-integrity angle we depend on. Applies a two-tier lens, checks known entries for movement, and appends to the living landscape register. Sibling of /tooling-scan (which scans for dev tooling that helps us build). |
Landscape Scan
Survey the field around CapableMind and keep the living landscape register current. This is the competitive/comparable-project scan — for tooling that could help us build, use /tooling-scan instead.
Principle
Research is important, but it must push us to make something better — not defeat us. (steward)
Generative, not anxious. We already know capability is commodifying and our differentiator is governance + epistemic integrity. So scanning for "is someone more capable" is a treadmill — the answer is always yes and it doesn't matter. The scan that earns its keep is narrower (Tier 2 below). Breadth without that lens is just anxiety; resist it.
§0 — Resolve the workstream lens
This skill is workstream-parametrized — one method, many lenses ("simple exterior, complex underneath"). Default workstream: capablemind.
If invoked as /landscape-scan <workstream> (e.g. studium-engine), FIRST read ~/_Dev/CapableMind-AI/docs/thinking/David/research/lens-<workstream>.md and use its values — the bet, what a real threat looks like (the Tier-2 focus), the register path, and workstream-specific sources — in place of the CapableMind specifics named below. Those specifics are the capablemind lens (also written out in lens-capablemind.md); the card overrides them per workstream. If no card exists for the named workstream, say so and stop — never scan against a guessed lens.
The two-tier lens
- Tier 1 — Capability / standards pulse (quick): what's becoming table-stakes, what techniques/standards are worth borrowing, what's worth a deeper look later. Skim.
- Tier 2 — The threat watch (the real work): is anyone building governed / constitutional / epistemic-integrity memory — memory you can trust, not just memory that's capable? That is the only landscape that can defeat our bet. Spend the weight here.
Movement matters as much as novelty. Re-check known register entries for change, not just hunt new ones. (Canonical lesson: Hermes's self-improvement loop read "overstated/static" on 2026-04-09 and "verified real" on 2026-05-27 — same name, moved underneath.)
Procedure
-
Open the register —
~/_Dev/CapableMind-AI/docs/thinking/David/research/landscape-register.md. Read current entries,Last scanned, and the watch list. This is the canonical source — append in place, never fork. -
Tier 2 sweep first (governed/epistemic-integrity memory). Targeted search — e.g. "constitutional AI memory", "governed agent memory", "trustworthy / auditable memory substrate", "honest degradation memory", "human-in-the-loop memory governance". The expected (and so-far-true) result is nothing credible — confirm that's still true, and name anything adjacent that's drifting toward the lane.
-
Tier 1 pulse. New capability/tooling/standards in the space (agent frameworks, memory systems, skill standards, inference infra). Sources: GitHub (trending + stars in the niche), HN, the register's watch list (review the highest-priority unreviewed item, e.g.
obra/superpowers), arXiv for the Tier-2 terms, and theagentskills.ioecosystem (the one real standardization signal). Also re-check known entries for movement. -
Assess each find through the lens. For each: what it is · overlap with CapableMind · governed? · verdict (
THREAT | LEARNING | CONFIRMS | TOOLING | INFRA). Verify from source/code or primary docs, not marketing — cite. Mark inferences[SPECULATIVE]. The governance/epistemic-integrity axis is the signal; capability rank is noise. -
Dedup + depth. Check the find isn't already in the register or a dated study (
research/). If a find is consequential enough to warrant detail, spin a dated deep-dive (e.g.research/YYYY-MM-DD-<name>-study.md) and link it from the register's index — don't bloat the register itself. -
Update the register. Append/adjust entries, update
Last scanned, move reviewed items off the watch list. Then surface a short digest to the steward: Tier-2 status (still open? / something moving?), notable new Tier-1 finds, and any movement in known entries. Recommend deeper scouts where warranted — but don't decide adoption: findings are for steward + jurist (governed initiative).
Output
A tight digest (not a wall):
- Tier 2: lane still open? / anything drifting toward it?
- New / moved (Tier 1): the few that matter, each one line with its verdict.
- Register updated: what changed; any new dated study spun.
- Recommend: any find worth a full scout, or a watch-list item to prioritize next time.
Constraints
- Generative, not anxious. The output should make us build better, not feel behind. If a scan only produces "X is more capable," it found nothing — re-aim at Tier 2.
- Verify from source. No claims from marketing pages or memory; read the repo/primary docs and cite. Mark speculation.
- Surface, don't decide. Adoption/architecture calls are steward + jurist territory (the loop is load-bearing). The scan informs; it does not commit.
- Proportion. This is a couple-of-days pulse, not a research project. The lens matters more than coverage; don't let the scan become the work instead of the building.
- Pool with Seb. He scans independently — the register is the shared canonical record so finds aren't double-chased.