dira decision memory for AI coding agents
$dira whyint-0003
int-0003Replace the tool pile — kazi + dira + a subset of skills is the whole stackactive 2026-07-30
int-0003Replace the tool pile — kazi + dira + a subset of skills is the whole stack
int-0003 ◆ active 2026-07-30

Replace the tool pile — kazi + dira + a subset of skills is the whole stack

Obsidian, Linear, and a drawer of half-adopted tools each failed at the same layer and were kept for the same reason: none of them owned direction, so none of them could be deleted. dira is only worth building if adopting it lets tools leave.

The discipline this imposes is subtractive, and it cuts against every instinct to add features: dira takes the direction layer and **nothing else**. Not tasks, not calendar, not journal, not notes-as-a-second-brain. Execution belongs to kazi; doneness belongs to kazi; scheduling belongs elsewhere. The closed entry set (cst-0002) is the structural guard on this intent — every proposed sixth entry kind is a proposal to become Obsidian.

Achieved when: at least two tools are uninstalled and not missed, and no session needs a surface outside Claude Code, kazi, and dira to answer "what are we doing and why."

**Sharpened 2026-07-30 (`dec-0013`).** "At least two tools" named no tool and no date, so nothing could falsify this intent. The first specific target is now `~/.claude/skills/sitrep` — a hand-executed version of `dira map` whose gather phase reads five sources and hand-glosses every id. It reduces to a judgement layer over `dira map` plus `gh`, or the tool-reduction thesis is weaker than claimed. Note the target is *reduction*, not deletion: the verdict, the weighted percentage, the risks and the trend are judgement and stay with the human.

Alternatives on record — 0

no alternatives recorded — an intent states a direction rather than choosing between options

Every decision in this repo is recorded like this one.

Keep your own →