Read phronemophobic "Treemaps are awesome!" and evaluated it against
the codebase-cartography skill (text-first, diff-able contract): verdict
— adopt proportional scale encoding (weight bar + Composition block),
reject literal treemap graphics (Mermaid treemap is v11.5+ ESM-only,
CORS-blocked under file://, incompatible with vendored v9.4.3 IIFE)
Rule-of-5 review of the proposal (rule-of-5-universal):
NEEDS_REVISION — 2 HIGH (weight semantics conflicted with the skill's
exclusion rule; no LOC collection command specified). Revised
accordingly
Implemented in incitaciones (commit 03f2492):
zoom-levels.md macro collect step 5 (LOC/file-count weight via
tokei/cloc + rg+wc fallback; two distinct metrics: Weight = analysis
code w/ exclusions, Composition = disk reality w/ excluded code),
10-cell ▓/░ bar spec, bulk-distribution fact, decision record;
report-template batch bullet; prompt-task v1.4.0, manifest 2026-09-05,
CHANGELOG [Unreleased]
Follow-up Rule-of-5 review of the committed state caught a CRITICAL:
the two-part edit to report-template.md was rejected atomically
(edits[0] never applied) yet the CHANGELOG claimed it had — macro mode
shipped unedited. Fixed in 7f080af + composition path/name
classification rule, tokei -e note, -0/-0
xargs hardening
All gates green (validate-distilled, sync-manifest, check-links,
lefthook); both commits pushed, main up to date
Next: release incitaciones (npm version patch + tag
→ CI publish), then reinstall skills
(pi install npm:incitaciones) so installed copies pick up
v1.4.0 content
Scaffolded PhormaSci/skills as a skill-set monorepo (npm workspaces,
pi.skills manifests, Agent Skills spec validator + scaffolder);
distribution via npx skills add PhormaSci/skills — npm
publishing dropped after catching my own conflation of the charly-vibes
npm login (that account belongs to incitaciones, not this repo)
Built + shipped the investigations set (commit
54c4de9): 8 inv-* skills split by
responsibility — gathering (hypothesis/experiment/evidence), tracking
(inv-renew/inv-next mirroring /renew, /next), extraction
(inv-conclusions/inv-close) — plus the inv-tools script: append-only
records.jsonl as single source of truth, schema-validated
appends, generated state.md/INDEX.md,
bujo-style dashboard
Design grounded in Meta's Organizational Second Brain + the LLM-Wiki
memory blueprint + bullet-journal orchestration; rationale recorded in
sets/investigations/research/design-rationale.md
(330fb95)
Rule-of-5 review of the set: READY after fixes
(12849cd) — 2 HIGH (repeated CLI flags overwrote
--artifact lists; evidence recorded against experiments was
invisible in hypothesis rollup and conflict detection), plus per-type
flag allowlists, frontmatter quote escaping,
INV_STALE_DAYS
Docs housekeeping (112c6ea): install flags corrected
against the actual skills CLI
(-a/-s comma-separated), phantom
~/.agents/skills claim removed, set README fences fixed,
CHANGELOG backfilled at 0.1.0
Dogfood:
investigations/parallel-agent-journal-conflicts opened
(H-1: parallel agent appends to records.jsonl merge cleanly in git) —
journal deliberately left uncommitted until its first experiment
Next: run H-1's experiment (delegated agents
appending from separate checkouts); per-agent staging files merged by
inv-renew is the fallback if conflicts confirmed
Session:17:54
(incitaciones — close-skill journal fix shipped + local install)
Continued from the treemap→cartography session: implemented the
journal-resolution fix for the close skill (bd
incitaciones-kmj)
Intermediate grep/sed resolver failed acceptance testing twice
(column-aligned inline comments survived sed strip;
JOURNAL_PATH="$JORNAL" evaluated empty —
JORNAL doesn't exist in clean shells); final design sources
.bashrc in a throwaway subshell (shellval())
so bash itself handles comments/quotes/nested vars — no parser, no
eval
Commits 0868ae0 + 53cfbd0 pushed;
verbatim-from-skill acceptance test in env -i is the gate
that caught both failures; ❌ close: [has frontmatter] in
validator is pre-existing (like next), not from these
edits
Ran ./install.sh --global from the checkout: installed
skills now carry the fix (close resolver + cartography
weight/composition) ahead of the npm release;
~/.config/agents/skills copies identical
Live-verified in this very session: non-interactive
/close resolved to canonical
~/para/areas/jornal + areas/log
Next: release incitaciones (npm patch + tag → CI
publish → reinstall from npm; bd incitaciones-8pv); check
pre-existing has frontmatter validator flags for
close/next