ground

Name: ground
Author: codeclawd

16stars6forks

How to install ground?

codeclawd/fable-mode/ground

$npx -y skills add codeclawd/fable-mode --skill ground

Installs into the current project.

Summary

Self-terminating grounding loop for a development step. Turns assertions, assumptions, and inferences into verified knowledge via an evidence-backed ledger; self-resolves code-determinable forks and logs them; escalates only genuine preference forks; spawns an independent verifie

Files · 1

View on GitHub

SKILL.md

1	# /ground — executable grounding protocol
2
3	Makes the "verify before done" discipline self-terminating (a ledger with a termination
4	test), self-resolving (a decision policy for forks), and independently verified (a cold
5	second agent), so it runs to completion without you re-prompting "ground further" each step.
6
7	## Core invariant
8
9	Do not return to the user while any load-bearing claim is UNVERIFIED or any code-determinable fork
10	is unresolved. "Looks done" is not a termination criterion — an empty UNVERIFIED column is.
11
12	## 1. Build the ledger
13
14	List every claim the step rests on, as a table: `\| # \| Claim \| Status \| Evidence \|`
15
16	- A row is VERIFIED only when Evidence is a concrete `file:line` (read) or a command **+ its
17	output** (run).
18	- "Structurally certain" / "by reading" / "likely" / "should" → UNVERIFIED. No exceptions.
19	- Behavioral claims — control flow, ordering, teardown, exception propagation, state
20	mutation, concurrency — require run evidence, not read evidence. Reason only to design the
21	check; then run it.
22	- For any library / framework / API claim, ground truth is the installed version or the official
23	docs (Context7) — not memory.
24	- A claim too vague to mark either way is split into checkable sub-claims.
25
26	## 2. The five moves
27
28	1. Read breadth-first — topology: imports, calls, where state lives (Grep/Glob/Read; fan out
29	parallel searches for fan-out questions).
30	2. Read the private internals — closures, captured vars, callbacks, indirection — not just exports.
31	3. Test each plan claim against the code — grep the counts, check the visibility.
32	4. Name what is intentionally excluded, and why.
33	5. Write findings into the plan / ledger.
34
35	## 3. The loop + termination
36
37	Run the moves, fill the ledger, repeat. Each pass must verify ≥1 row or split a vague claim.
38	Terminate when a full pass yields no new verifications AND the UNVERIFIED column is empty for
39	all load-bearing claims.
40
41	## 4. Decision policy — make forks resolve themselves
42
43	For each fork the grounding surfaces, classify:
44
45	- Code-determinable — one branch is correct given how the code wires, a project invariant, or
46	the development trajectory. → Decide it. Write *decision + grounded reasoning + rejected
47	alternative(s)* into a decision log. Do not ask.
48	- Preference / value — needs the user's priorities, risk tolerance, or product direction; or
49	is irreversible and expensive. → Stop and surface it inline (markdown list, never
50	AskUserQuestion) with a grounded recommendation.
51	- Unsure which class → treat as preference and surface. Conservative on purpose.
52
53	The bar: ask only what the code genuinely cannot answer.
54
55	## 5. Independent verification
56
57	When the ledger is clean, spawn the grounding-verifier agent (cold, read-only). Give it the
58	step goal + the ledger + the diff (if any) — not your justification narrative. It performs:
59
60	- (a) Row-check — each cited evidence actually supports its claim; re-run behavioral checks
61	(the run is ground truth regardless of reader).
62	- (b) Gap-scan — independently hunt load-bearing claims the ledger omits: bypass paths,
63	untested assumptions, unnamed exclusions, count claims that are off.
64	- (c) Summary-fidelity — confirm the §6 packet faithfully describes the diff.
65
66	Resolve every REFUTED / INSUFFICIENT finding (re-ground, fix, or escalate) before surfacing.
67	The verifier reduces but does not eliminate error — it breaks the self-grading loop. Independence
68	comes from cold context + adversarial stance, not model diversity (same model is fine). For code
69	changes you may also run the bundled `/code-review` and `/verify` skills alongside it.
70
71	## 6. The review packet — abstraction parity
72
73	The ledger is an audit trail, not a review surface. Produce a separate packet at **the
74	altitude of the user's original request**, descending to code only in an appendix:
75
76	1. The ask — restated in the user's words.
77	2. What I did — plain English, same altitude.
78	3. Decisions + why + rejected alternatives — one line each, plain English.
79	4. Genuine forks for the user, if any.
80	5. One-command behavioral proof — before/after, a single command the user can run.
81	6. Appendix — ledger, decision log, full diff. There for drill-down, never required to review.
82
83	## 7. Ceiling (autonomy)
84
85	Default ceiling 2: ground → decide code-determinable forks → implement → verify → **stage
86	everything and stop before commit** for sign-off. Ceiling 1: also commit autonomously. Ceiling 3:
87	widen

People who liked this also viewed

6 picks

Type

ground

16stars6forks

View on GitHub

How to install ground?

codeclawd/fable-mode/ground

$npx -y skills add codeclawd/fable-mode --skill ground

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# /ground — executable grounding protocol
2
3	Makes the "verify before done" discipline self-terminating (a ledger with a termination
4	test), self-resolving (a decision policy for forks), and independently verified (a cold
5	second agent), so it runs to completion without you re-prompting "ground further" each step.
6
7	## Core invariant
8
9	Do not return to the user while any load-bearing claim is UNVERIFIED or any code-determinable fork
10	is unresolved. "Looks done" is not a termination criterion — an empty UNVERIFIED column is.
11
12	## 1. Build the ledger
13
14	List every claim the step rests on, as a table: `\| # \| Claim \| Status \| Evidence \|`
15
16	- A row is VERIFIED only when Evidence is a concrete `file:line` (read) or a command **+ its
17	output** (run).
18	- "Structurally certain" / "by reading" / "likely" / "should" → UNVERIFIED. No exceptions.
19	- Behavioral claims — control flow, ordering, teardown, exception propagation, state
20	mutation, concurrency — require run evidence, not read evidence. Reason only to design the
21	check; then run it.
22	- For any library / framework / API claim, ground truth is the installed version or the official
23	docs (Context7) — not memory.
24	- A claim too vague to mark either way is split into checkable sub-claims.
25
26	## 2. The five moves
27
28	1. Read breadth-first — topology: imports, calls, where state lives (Grep/Glob/Read; fan out
29	parallel searches for fan-out questions).
30	2. Read the private internals — closures, captured vars, callbacks, indirection — not just exports.
31	3. Test each plan claim against the code — grep the counts, check the visibility.
32	4. Name what is intentionally excluded, and why.
33	5. Write findings into the plan / ledger.
34
35	## 3. The loop + termination
36
37	Run the moves, fill the ledger, repeat. Each pass must verify ≥1 row or split a vague claim.
38	Terminate when a full pass yields no new verifications AND the UNVERIFIED column is empty for
39	all load-bearing claims.
40
41	## 4. Decision policy — make forks resolve themselves
42
43	For each fork the grounding surfaces, classify:
44
45	- Code-determinable — one branch is correct given how the code wires, a project invariant, or
46	the development trajectory. → Decide it. Write *decision + grounded reasoning + rejected
47	alternative(s)* into a decision log. Do not ask.
48	- Preference / value — needs the user's priorities, risk tolerance, or product direction; or
49	is irreversible and expensive. → Stop and surface it inline (markdown list, never
50	AskUserQuestion) with a grounded recommendation.
51	- Unsure which class → treat as preference and surface. Conservative on purpose.
52
53	The bar: ask only what the code genuinely cannot answer.
54
55	## 5. Independent verification
56
57	When the ledger is clean, spawn the grounding-verifier agent (cold, read-only). Give it the
58	step goal + the ledger + the diff (if any) — not your justification narrative. It performs:
59
60	- (a) Row-check — each cited evidence actually supports its claim; re-run behavioral checks
61	(the run is ground truth regardless of reader).
62	- (b) Gap-scan — independently hunt load-bearing claims the ledger omits: bypass paths,
63	untested assumptions, unnamed exclusions, count claims that are off.
64	- (c) Summary-fidelity — confirm the §6 packet faithfully describes the diff.
65
66	Resolve every REFUTED / INSUFFICIENT finding (re-ground, fix, or escalate) before surfacing.
67	The verifier reduces but does not eliminate error — it breaks the self-grading loop. Independence
68	comes from cold context + adversarial stance, not model diversity (same model is fine). For code
69	changes you may also run the bundled `/code-review` and `/verify` skills alongside it.
70
71	## 6. The review packet — abstraction parity
72
73	The ledger is an audit trail, not a review surface. Produce a separate packet at **the
74	altitude of the user's original request**, descending to code only in an appendix:
75
76	1. The ask — restated in the user's words.
77	2. What I did — plain English, same altitude.
78	3. Decisions + why + rejected alternatives — one line each, plain English.
79	4. Genuine forks for the user, if any.
80	5. One-command behavioral proof — before/after, a single command the user can run.
81	6. Appendix — ledger, decision log, full diff. There for drill-down, never required to review.
82
83	## 7. Ceiling (autonomy)
84
85	Default ceiling 2: ground → decide code-determinable forks → implement → verify → **stage
86	everything and stop before commit** for sign-off. Ceiling 1: also commit autonomously. Ceiling 3:
87	widen

People who liked this also viewed

6 picks

Type