UNPKG

paqad-ai

Version:

Spec-driven development framework — AI agents that think before they type

github.com/Eliyce/paqad-ai

Eliyce/paqad-ai

65 lines (49 loc) • 3.99 kB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
--- name: rule-editor description: "Adds, edits, removes or downgrades one rule in rules-as-scripts with a per-rule cascade and no global rebuild: mints or keeps the stable RL-hash id, regenerates only that rule's scripts, and writes the map through src/rule-scripts/apply.ts. Use when the user asks to add, edit or remove a rule, or to mark one unverifiable." metadata: model_tier: medium cacheable: 'false' output_format: markdown paqad-meta: skill.meta.yaml --- ## What It Does Owns the mutation cascade so adding, editing, or removing a rule recomputes only what changed: never a full regeneration: | Mode | Effect | | --------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- | | add | Inserts the bullet into the target rule file, mints a stable `RL-<hash>` id, then classifies + generates scripts for **only the new rule**. | | edit | Updates the bullet text, **preserves the id**, regenerates only that rule's scripts, and surfaces a behavior diff via Decision Pause when scripts change. | | remove | Removes the bullet, moves the map entry to `archived:` (delayed delete); scripts pruned on the next regen pass. | | downgrade | Keeps the rule, deletes its scripts, flips the map entry to `verifiability.kind: unverifiable` with a recorded reason. | ## Use This When - A user prompt matches the priority-225 router rules `add rule`, `edit rule`, `remove rule`, or `mark rule as unverifiable`. ## Inputs - The target rule file (for `add`) or the `RL-<id>` (for edit/remove/downgrade). - The new rule text (`add`, `edit`) or downgrade reason (`downgrade`). - `docs/instructions/rules/rule-script-map.yml` (written only via the bundled wrapper → `src/rule-scripts/apply.ts`). ## Procedure 1. Apply the deterministic mechanics: ``` node scripts/edit-rule.mjs add <project-root> <source-rel> "<text>" node scripts/edit-rule.mjs edit <project-root> RL-<id> "<text>" node scripts/edit-rule.mjs remove <project-root> RL-<id> node scripts/edit-rule.mjs downgrade <project-root> RL-<id> "<reason>" ``` 2. For `add` and `edit`: re-run `analyze rules` (idempotent, touches only the changed rule) then `generate rule scripts` with `only_rule_id` set to the affected id, so only that rule's scripts are regenerated. 3. For `edit`, diff the regenerated scripts against the prior ones; if behavior changed, surface a Decision Pause packet before accepting. 4. Present a `## Rule Change Summary` naming the id, the mode, and any scripts added/removed. ## Output Contract - The rule markdown updated with a preserved/minted `<!-- @rule RL-<hash> -->` marker. - `rule-script-map.yml` updated via `src/rule-scripts/apply.ts` (archive on remove, unverifiable on downgrade, script set on add/edit). - A `## Rule Change Summary` markdown block. ## Escalate / Stop Conditions - `edit` / `remove` / `downgrade` on an id not found on disk stops with a clear error: never guesses which rule was meant. - Behavior-changing `edit` regenerations surface a Decision Pause diff before acceptance. - Never edit `rule-script-map.yml` by hand; the wrapper is the only path. - A new rule that contradicts an existing one is surfaced as a conflict (via `rule-analyzer`) before its scripts are written. ## Resources - `references/edit-modes.md`: the four sub-modes + cascade contract. - `runtime/base/skills/rule-editor/agents/openai.yaml`: agent interface metadata. - `src/rule-scripts/editor.ts`: markdown add/edit/remove mechanics. - `src/rule-scripts/mutate.ts`: per-rule map mutations. - the Decision Pause Contract (in the framework bootstrap): packet semantics for edit diffs + conflicts.