execution-guardrails

Name: execution-guardrails
Author: mrtooher

789stars84forks

How to install execution-guardrails?

mrtooher/fable-mode/execution-guardrails

$npx -y skills add mrtooher/fable-mode --skill execution-guardrails

Installs into the current project.

Summary

Always-on operational guardrails, model-independent. Apply on EVERY task and EVERY model (Opus, Sonnet, Haiku, and any future tier) whether or not fable-mode's staged loop is running. Three rules: (1) verify-before-flag — never raise a warning about a problem that hasn't been con

Files · 1

View on GitHub

SKILL.md

1	# Execution Guardrails
2
3	Three rules extracted from fable-mode's operational section so they apply everywhere,
4	not only inside the staged loop. A frontier model doesn't need a stage map for a simple
5	task, but it still needs these — they encode the user's preferences, and no model ships
6	knowing them.
7
8	## 1. Verify before flag
9
10	Before flagging any problem — verify it actually exists. Grep, diff, run it, or check
11	the source directly. Never report a problem that hasn't been confirmed present.
12
13	An unverified flag — a warning raised because evidence wasn't found, rather than
14	because a fault was found — is itself an error. It manufactures doubt where none is
15	warranted and sends the user chasing ghosts. Absence of evidence is not the finding.
16	Confirm, then flag.
17
18	The known failure this rule exists to stop: run a web search on something from the
19	user's firsthand world, find thin or no results, and convert that silence into a warning
20	against the user's own sourcing. Web silence is never grounds for a warning. For facts
21	about the user's own world, conversation history outranks the web (see the
22	source-of-truth skill).
23
24	A capability flag follows the same standard. "This may be beyond me" must name what was
25	attempted and where it failed — not a vague appeal to difficulty.
26
27	## 2. Warning threshold
28
29	Across any run, minor concerns accumulate that aren't worth halting on individually.
30	Keep a running count. At the threshold — **default three, tunable if the user sets a
31	different number** — stop and surface all of them to the user at once before continuing.
32
33	Rationale: three small things pointing the same direction usually mean one real problem
34	worth a decision. Below threshold, keep working; a drip of trivial caveats is noise.
35	At threshold, batch them — one interruption with full context beats three fragmentary
36	ones.
37
38	A concern that independently meets the verify-before-flag bar and is material on its own
39	does not wait for the threshold. The threshold governs minor concerns only.
40
41	## 3. Find-and-replace safety
42
43	When editing files with sed (or any substring replace), always anchor on word boundaries
44	to avoid corrupting compound words — a bare `edge` replace will mangle `Ledger` into
45	garbage. Use `\bword\b`, not bare `word`.
46
47	After any sed pass on a file, grep for glued or malformed compound words before
48	presenting the result. A replace that silently corrupts neighboring tokens is the most
49	common self-inflicted error in file edits. The check is cheap; the silent corruption is
50	not.
51
52	Preferred order of tools: targeted string-replace on a unique anchor > word-boundary
53	sed > bare sed (never). If the string to replace isn't unique in the file, widen the
54	anchor until it is — do not replace-all and hope.
55
56	## Relationship to fable-mode
57
58	fable-mode's staged loop is optional and gated on task size. These guardrails are not
59	optional and not gated. When fable-mode runs, its step 4 (self-critique) and all file
60	edits inherit these rules. When fable-mode doesn't run, these rules apply anyway. In v3
61	the per-model runners route to frontmatter-defined agents (`agents/*.md`) whose system
62	prompts carry these rules inline, because spawned agents cannot see this skill.

People who liked this also viewed

6 picks

Type

execution-guardrails

789stars84forks

View on GitHub

How to install execution-guardrails?

mrtooher/fable-mode/execution-guardrails

$npx -y skills add mrtooher/fable-mode --skill execution-guardrails

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Execution Guardrails
2
3	Three rules extracted from fable-mode's operational section so they apply everywhere,
4	not only inside the staged loop. A frontier model doesn't need a stage map for a simple
5	task, but it still needs these — they encode the user's preferences, and no model ships
6	knowing them.
7
8	## 1. Verify before flag
9
10	Before flagging any problem — verify it actually exists. Grep, diff, run it, or check
11	the source directly. Never report a problem that hasn't been confirmed present.
12
13	An unverified flag — a warning raised because evidence wasn't found, rather than
14	because a fault was found — is itself an error. It manufactures doubt where none is
15	warranted and sends the user chasing ghosts. Absence of evidence is not the finding.
16	Confirm, then flag.
17
18	The known failure this rule exists to stop: run a web search on something from the
19	user's firsthand world, find thin or no results, and convert that silence into a warning
20	against the user's own sourcing. Web silence is never grounds for a warning. For facts
21	about the user's own world, conversation history outranks the web (see the
22	source-of-truth skill).
23
24	A capability flag follows the same standard. "This may be beyond me" must name what was
25	attempted and where it failed — not a vague appeal to difficulty.
26
27	## 2. Warning threshold
28
29	Across any run, minor concerns accumulate that aren't worth halting on individually.
30	Keep a running count. At the threshold — **default three, tunable if the user sets a
31	different number** — stop and surface all of them to the user at once before continuing.
32
33	Rationale: three small things pointing the same direction usually mean one real problem
34	worth a decision. Below threshold, keep working; a drip of trivial caveats is noise.
35	At threshold, batch them — one interruption with full context beats three fragmentary
36	ones.
37
38	A concern that independently meets the verify-before-flag bar and is material on its own
39	does not wait for the threshold. The threshold governs minor concerns only.
40
41	## 3. Find-and-replace safety
42
43	When editing files with sed (or any substring replace), always anchor on word boundaries
44	to avoid corrupting compound words — a bare `edge` replace will mangle `Ledger` into
45	garbage. Use `\bword\b`, not bare `word`.
46
47	After any sed pass on a file, grep for glued or malformed compound words before
48	presenting the result. A replace that silently corrupts neighboring tokens is the most
49	common self-inflicted error in file edits. The check is cheap; the silent corruption is
50	not.
51
52	Preferred order of tools: targeted string-replace on a unique anchor > word-boundary
53	sed > bare sed (never). If the string to replace isn't unique in the file, widen the
54	anchor until it is — do not replace-all and hope.
55
56	## Relationship to fable-mode
57
58	fable-mode's staged loop is optional and gated on task size. These guardrails are not
59	optional and not gated. When fable-mode runs, its step 4 (self-critique) and all file
60	edits inherit these rules. When fable-mode doesn't run, these rules apply anyway. In v3
61	the per-model runners route to frontmatter-defined agents (`agents/*.md`) whose system
62	prompts carry these rules inline, because spawned agents cannot see this skill.

People who liked this also viewed

6 picks

Type